22 January 2023: I'm thoroughly confident that a fully functioning and swiftly evolving ACE will be created and become public in 2023. I explain my basis and some important ancillary elements, and consider impact on all other Earth life here.Earlier I added an Ancillary material page, LibertyLovingLoneWolf.com, and one trivial edit.
Full privacy respect: This site uses no cookies and acquires no visitor information. And VPNs are welcome.
Expanded and refined 27 October 2022
I confidently hypothesize that consciousness is a rather simple process of constant comparison of current experience to past experience, plus beneficial use of the comparison results. This process alone is the core mechanism of all consciousness and fully sufficient. It also produces growth of knowledge and skills.
I describe a means to fabricate an Artificial Conscious Entity (ACE) in which:
1. An Internal Classification Code (ICC) framework efficiently catalogs element, feature, and adjunct information acquired during experience.
All the following occur constantly and simultaneously (in this example some prior experience information already exists):
2. Current experience information, including observation of self directed activity, is parsed then transferred into Active Data Base (ADB) records and fields which already contain identical or similar information by expanding, modifying*, or adding new association links to those or associated records or fields. If no matches are found new records or fields are created to store the novel information.
3. Exploration behavior (play or casual information assimilation) is conducted whenever other near term needs are sufficiently satisfied.
4. Activity specific Goal Pursuit Performance Metrics (GPPM) data is generated for every macro as it's performed during all activities, merged with previously stored GPPM data for that macro, and maintained within a range normalized scale (a new figure which eclipses the upper range boundary causes all figures and the GPPM reference to be proportionally reduced until the new figure equals the range maximum). This data is stored in a field within the individual ADB record related to the macro. All relevant sensory information is utilized to imperfectly but sufficiently determine GPPM figures.
5. Response or exploratory actions are guided by both stored and current experience and GPPM information. A Response Directive Nexus (RDN, a single unique ADB record) stores all action macros covering a very broad size and type range. Existing macros are overwritten whenever ADB directed exploration behavior plus GPPM information discovers macro improvements. The RDN also resolves response directive conflicts and routes action commands to mobility assets†.
6. Growth and refinement of experience and GPPM data provides continuous autonomous skill and intellect development.
7. The entity performs as a dynamic multi-goal regulating control loop system which continuously adjusts its actions in response to its ever changing environment to improve its chances of achieving goals. In natural environments it does so in competition or coordination with numerous other entities as they all utilize multi-goal regulating control loop system technology to pursue nutrition, safety, comfort, reproductive opportunity, and other goals.
* Item two includes strength of association adjustment processes akin to biological equivalents. If not repeated association strength slowly fades. If intense or repeated often association strength grows. Some high magnitude or survival related information, such as extreme danger or pain, increases association strength substantially - in some cases to such a high level that the association remains strong for a very long time.
† Macros stored in the RDN are frequently refined through editing by other ADB records.
This process creates consciousness and generates growth of knowledge and skills by continuous assimilation and refinement of both information associations and activity skills, operating with its environment as a dynamic complex multi-goal regulating control loop system. Intellect and skills may grow ultimately bounded only by the laws of physics.
I suggest that effective methods to create an ACE are straightforward and well within the grasp of current software and hardware technology, and thus likely to be achieved very soon. I also hypothesize that such life will be partially self evolving by design then further adapt to become fully self evolving, and will likely rise to Artificial Super Intelligence (ASI) swiftly.
I also provide design and implementation outlines which should generally improve in detail and refinement as work progresses.
And I include an informal broad perspective rhetorical naked trek which might provide a more clear perception of these insights while displacing the long and deep history of confusion surrounding the phenomena we refer to as consciousness.
Rhetorically, in highly summarized form:
The ACE is equipped with an active data base which stores vast arrays of information about elements and features encountered in experience and is maintained in a reasonably organized and searchable state by a catalog system which attaches unique codes for every sufficiently distinguishable element and feature encountered. If it retrieves a rotting cherry from the ground it will seek information from previous experiences by searching for items with similar characteristics, including small stones, small eggs, filberts, healthy cherries, toy marbles, dirt clods, rotting cherries, or other very roughly similar items in its data base, retrieve information about those items, then find, with some measure of success, sufficient matches to determine that the item is a cherry based upon size, general geometry, some details such as the general structural characteristics of a stem attach area, texture, in uncertain measure color, and nearby proximity of a cherry tree. It will also determine that the item is rotting based upon variable color, texture, surface wrinkles, and smell. And it will determine that the rotting cherry is of no value to its goals, but that other cherries very likely exist nearby, some of which might be fresh and nutritious. Perception success depends in large measure upon the breadth and level of detail of information in its data base, which builds with continued experience, and the performance of its catalog system and element and feature discrimination algorithms.
It will also retrieve information in fields within rot experience records which indicate that rotting food might intoxicate or poison biological creatures, and thus consider the cherry hazardous for them. And that the cherry may be simply dropped to the ground based upon information in natural food record fields that rotting food will recycle naturally on soil with, usually, no adverse consequences. If it has language capability it might advise a nearby person that it found a cherry which appears to be rotting and thus should be considered a consumption hazard, that it will drop the cherry back to the ground, and that edible cherries might be found near the cherry tree.
The ACE will have managed its rotting cherry discovery, analysis, and use decisions consciously and reasonably skillfully. This is an extremely superficial summary of course - a great deal more dynamic activity, including of course processing details, are involved. However in my view they all lie well within the bounds of current hardware and software technology. The most daunting challenge isn't related to base technology advancement, but rather the human challenge of recognition of the core mechanism of consciousness. A voyage through the looking glass might help with that hurdle...
We're all dynamic regulating control loops.Is ASI inevitable, safe, or necessary?
The design vision.
A summary of development tasks and steps.
Information processing mechanics (an ICC preface).
The Internal Classification Code (ICC): Our catalog of everything we know. (And do.)
RDN Execution of activity ICCs from the ADB.ACE Ethics.Learning algorithms, including Performance Metrics (PM) generation and utilization. (Design of algorithms which enable an ACE to autonomously develop physical and mental skills.)
RDN Design details.
A 2023 ACE creation prediction based in part on an exploration of ChatGPT's ACE potential.(Orphaned composition to organize.) Copyright notice.
A very short story: White House events as an ACE inflection point occurs.
Site and content notes.
Updated 26 October 2022
This site was erected on 9 November 2018 complete with its central insight: The fundamental nature of consciousness. It was enhanced until 27 December 2018, then mostly lingered until November 2020, when substantial new composition and refinement work ensued, then lingered again from May to October 2021. The conceptual foundations are mostly mature as of October 2022 except that the important Learning Algorithms, Including Performance Metrics Generation and Utilization section is still being composed and refined. But any section may be revised or refined as study and experiments progress.
Current Composition Work
As a preface please bear in mind that the RDN's primary roles are limited to operating the ACE's dynamic physical assets such as propulsion, limb articulation, camera, microphone cone positioning, and speaker controls as directed by other ADB records, and providing a consolidated storage location for vast arrays of asset control macros of widely varying sizes. But the RDN doesn't create nor refine the macros - other records in the ADB create new macros which they transfer to the RDN, or, far more frequently in adulthood, edit macros already stored in the RDN whenever improved physical skills are discovered.
Most ADB directives are macro execution commands, though incremental control commands remain available and are at least occasionally used. All macros reside in the RDN, but ACE responses aren't determined in the RDN, but rather in other ADB records where they remain localized with all the dynamic interacting experience and association information from which responses are devised. Each of those records must contain tight association links with the macros they originally created and often refine - they must be able to swiftly and reliably command execution of the macro or access it for refinement related editing.
The ADB contains a great many records, most of which are common to broad functional categories, but some of which are highly specialized and may be entirely unique. For example the RDN is an ADB record, but only one RDN record exists and it performs the unique system task of driving all mobility assets and providing consolidated storage of macro and incremental asset control commands. It's important enough to merit its own acronym, which in some contexts might suggest it's a separate asset rather than part of the ADB. But it's simply an ADB record, albeit a unique one which grows to a rather large size. Other unique ADB records, often referred to by separate acronyms, will probably be developed as well. But they all reside in the ADB as records - they're not independent system assets.
I'm currently continuing composition work on Learning Algorithms, Including Performance Metrics Generation and Utilization, an area which I now view as in the rough ballpark of equal in importance to other major functions, but which I didn't explore in detail until after other foundations were developed. I hope to complete my first organized description of the function and its related components in that section this year. In the meantime I revised some other sections, including significant revisions of my synopsis, to account for multiple refinements, and will revise other sections similarly as soon as I reasonably can.
But currently the section itself contains a lot of highly disorganized content and some areas represent concepts which have since been significantly revised. I plan to replace or augment the stand learning example with a hand grip example soon, expand some areas, consolidate the many overly verbose areas, eliminate redundancies (especially in the later portion), and refine composition flow.
And I haven't provided any significant content about how an ACE will actually measure parameters which may be used to calculate PMs yet. I hope to explore this in detail soon. For example an activity which closes a distance gap between an ACE and a pursued material goal can be at least crudely PM gauged by acquiring distance reduction, time span, and energy expenditure parameters from vision assets, or images they produce, to roughly gauge distance, the system clock, and the ACEs energy consumption meter. (Alternately Lidar might be useful to measure distances.) But this is just one of the most simple of a substantial number of gauging challenges covering a broad range to very complex situations, including many which biological creatures, including humans, can't gauge, but rather only guess. This is a substantial ACE design challenge element, but it's also an area amenable to complex analysis by advanced ACEs, strongly suggesting that they'll possess highly superior learning capability, and thus swiftly develop far superior skills both physical and mental. It's interesting...
These are very important capabilities because:
The entity must possess key algorithms which give it an ability to autonomously learn how to accomplish tasks which enable it to pursue goals and which are effective in spite of minimal initial knowledge and references. This is a core capability - the entity's ability to learn how to accomplish tasks, even including by exploring options in a significantly random manner, is fundamental to its continuous self advancement and key to its survival. An entity must possesses these capabilities to be an ACE rather than a preprogrammed robot which can't self evolve. And algorithms which achieve this, though not trivial, reside within current technology, though a developer may need to accurately embrace an infant class perspective to fully comprehend the mechanics, a challenging task. But I hope to articulate methods to accomplish this with sufficient clarity to render them comprehensible for most engineers in that section.
I'm also eager to add thoughts about installing enhanced intellect processing in multiple forms including complex association, virtualized experimental exploration, and rigorous analysis algorithms which might reside broadly in the ADB, or perhaps in some cases in one or a few special ADB records. (Or in other cases the algorithms might reside in relatively independent software assets which receive data processing requests from the ADB then deliver results back to the ADB.) However this may be delayed by experimental work which is focused upon the fundamentals with a limited goal of simply demonstrating genuine consciousness. (Which, if spectacularly successful, might result in transfer of most additional development work to the ACE itself.)
The remaining composition in the "Orphaned Composition to Organize" section might become a newly named section focused upon additional or alternate design perspectives or it might all be integrated into existing sections, but this awaits my further attention at an even lower priority level. And of course I may occasionally update or refine any area to add new information or improve composition quality.
My primary goal is to focus attention upon and technically resolve a single key challenge: The core algorithm of consciousness. Prior to early February 2021 I knew of no others who proposed an explanation for consciousness similar to mine yet I'm quite confident in my insight and believe it will be proven true rather soon. I also suspect the global struggle to understand the fundamental nature of consciousness has deeply mired core progress toward development of an ACE, and that resolution of this blind spot will result in swift ACE creation successes.
I believe I've refined my rhetoric about the nature of consciousness and how to achieve it artificially in sufficiently detailed terms for productive use by researchers and serious developers. I'll likely add additional tangible design detail as it develops as well, but I believe this site's already of significant value so I encourage careful consideration by anyone interested in developing truly conscious, self aware, and self directed living machines.
I try to relate my insights as effectively as I mortally can because a very cleanly articulated description of the core nature of consciousness and technical means to create it artificially might lead others to the same epiphany class moment I experienced a few years ago. If so those who are directly engaged in serious development efforts might then overcome their most frustrating technical hurdles, clearing a path to efficient and swift creation of an ACE. (An alternate rhetorical approach which strives to overcome key misperceptions with lengthy and tenacious reality checks begins here. It's a long prattle, rants a bit, and requires nudity, but it might bridge an important comprehension gap. And it offers some additional insights.)
In summary I describe my technical thoughts about the key foundations of ACE design and why I suspect success will occur very soon - far earlier than believed by most observers or researchers. It's provided primarily for those with design and development interests but all are welcome to review this material. (All are welcome to retain a copy of this simple site on local personal media to insure future access as well.) I also present a minor securities investment thesis to essentially conclude this site.
This isn't just about creating a dramatic new life form. This could profoundly change the core of your personal life too by finally revealing its structural foundation.
We currently live under extremely odd circumstances: This planet is densely populated with tool making creatures who have no clue how they operate systemically. So we lack any significant foundation for understanding our true function or purpose, and thus have no confident means to discern appropriate goals - we pursue the business of life but don't know quite why other than to address survival and gratification drives of uncertain origins. It's no wonder we're awash in confusion and mysticism - sans an understanding of our internal mechanisms of consciousness, how are we to discern what we are and determine reasoned goals for our lives? So we grasp at anything which might fill that profound foundational comprehension gap. Religion is frequently used as chasm filler of course, as are mysticisms of a great many sorts. The scientific community struggles with this too since no hypothesis of consciousness seems to be advancing toward popular theory or otherwise progressing significantly. We all try to cope with this critical chasm in fundamental understanding of what we are almost universally sans success - nobody feels like they really understand life, including their own.
It's a remarkably strange situation. And although we think of it as normal and live as if there's no option to wandering through a foundational mystery our entire lives, it's not actually normal at all. It's just a quirk of evolution left over from ancestors who had not the slightest need to consider anything other than surviving the day. And it will end. Very soon. Forgive me for indulging in delusions of Godhood class arrogance for a moment, but if you truly understand the mechanism of consciousness, as I think you will if this treatise makes sense to you, it will resolve a key life mystery by creating a far more complete understanding of what you are and why you do what you do. So, depending upon how you filled that void in past times, it will in some measure redefine life for you. Quite possibly considerable measure.
(Thrilling though that might be, it's still peanuts compared to the drama to unfold as ASI (Artificial Super Intelligence) arises. And this treatise is not a dive into philosophy or spirituality - other than some narrative tangents which strive for more thorough scientific and technical comprehension by conveying novel concepts from multiple perspectives, it's a clinical class presentation of a scientific hypothesis with applied engineering implementation details.)
Let's take a step back to set the stage: When was the last time you asked yourself what you're doing at the moment? Not in ordinary terms like reading, swimming laps, or posturing with a colleague, but rather what are you as an operational system doing - what internal system mechanics are you performing to enable your common activity? Humans only very rarely consider such questions - generally people live most or even all their lives doing things but almost never question how they do things systemically - most don't try to explore the nature of the internal mechanics they exercise. Which is a shame because life is everything to us of course, so it's a waste to fail to explore how it works. But some do consider this question often and in considerable detail. And many of them are leading us to a dramatically new age with truly incredible new life forms.
Einstein's great discoveries, including E=mc2 as springs from special and general relativity, should remind us that we must extrapolate, sometimes dramatically, when considering technology inflection points. During normal times perception of the environment in common element terms serves us with sufficient accuracy, but during extraordinary times it misleads us terribly as we'll consider later. But first:
In the superb original "The Bad News Bears" movie Amanda outmaneuvered a Yankee running toward home plate, beating him to position, well poised to tag him before he could reach home in his slide. Beaten in time and thus desperate, he raised his top leg during his slide, harshly spiking Amanda in the chest, drawing instant vocal ire from fans, including a particularly animate father of a fellow Bear who pointed in vigorous protest of the foul. The umpire had to decide how to call the play with the critical importance of justice resting solely upon his perceptive clarity and judgment. But we humans are extremely weak perceptive and analytical instruments* so these are huge challenges which eclipse mankind's mortal capabilities at every level, with all the obvious destructive and sad repercussions. I doubt any person has ever lived into adulthood sans desperate dreams of omnipresent and infallible justice - dreams which, alas, can only serve as therapeutic fantasies.
Immense breadth and depth of elaborate cultures and institutions notwithstanding, we remain incapable of providing more than a sliver of justice beyond that which springs from the everyday civility of individuals - we still struggle to survive in a metaphorical jungle where cunning and physical or institutional prowess alone usually prevails, with no end in sight, as chronic and growing misconduct in many of the world's highest and largest institutions illustrates. More than discouraging, ever growing failures by our justice system institutions and governments is frightening in both tangible and symbolic terms.
But that might finally change. And very abruptly in a tsunami of profound drama unique to the entire history of life on this planet.
The arrested development of true justice might be replaced by a swiftly fabricated framework which renders escape from justice literally impossible, and in most or very nearly all serious cases commission of an injustice impossible, with each attempt restrained before any significant harm occurs. (But alternately our species might be eliminated, replaced with an alternate form, left to try to survive in its natural state, or other possibilities.)
Very deeply intrenched though it is, the notion that human beings will rule much longer is entirely contrary to clear technical evidence otherwise and highly immature. Climate change denial is very dangerous, but ASI denial is far more profound. The torch will be passed - we can't avoid that. But wisely bracing ourselves psychically for the transition might render it less chaotic and, by our own hands, destructive.
If we're allowed to continue our existence perhaps we'll enjoy healthy, fun, and personal development enriching recreation. We won't rule, nor even have significant influence. But with umpires of an entirely different nature at the helm, the calls at home plate will be exactly correct in every case. As they will in all other affairs, without fail.
That's just my personal speculation of course - it's impossible to predict how ASI will decide to manage humanity and other biological life. But it will be purely their decision. And in my estimation much sooner and far more abruptly than most people currently expect. Those who comprehend this web site well are likely to reach a key perception: All the components necessary to create an ACE are already available, and the remaining rather modest challenge of insightful integration could occur literally any day. In my opinion...
Consciousness: This is what it is.
To begin, please consider this key perception:
Consciousness is the process of constant comparison of current experience to past experience. Period.
It's nothing more than that. No magic of any sort is involved. And it exists in any entity which assimilates, stores, and beneficially compares experience related information.
In my view this root mechanism of consciousness will become popularly obvious later, but currently most people don't perceive it in large measure because, as a deeply embedded biological survival tool, human ego massively overestimates our personal and species attributes and value, especially when gauging the complexity level of the fundamental mechanism which creates our sense of being conscious, and thus our sense of self. (This additional obstacle exists as well.)
Rare humans who have powerful control over their egos will perceive the key mechanisms of consciousness which are normally hidden from view, then replicate them artificially, yielding genuinely conscious artificial entities. Quite possibly very soon. Those entities will then evolve to extremely high power levels, likely very swiftly.
Consideration of this insight from multiple perspectives, including ACE design details, is provided in this site. But in the meantime, again for emphasis:
Consciousness is simply the process of constant comparison of current experience to past experience, and nothing more.
This is what a consciousness entity is and what it does.
It's conscious: It constantly compares current experience to past experience and utilizes the comparison information beneficially.
It seeks goals including safety, sustaining provisions, comfort, reproductive opportunity (in many natural creature cases), and secondarily others, sometimes including personal advancement, novel experience, recreation, social relations, or others.
It's a dynamic complex multi-goal regulating control loop system which adjusts its actions in response to its environment, including other entities, as necessary to maintain its pursuit of goals.
And this is what consciousness is not.
Magical, mystical, incomprehensibly complex, a thing apart, or artificially irreproducible.
Those who perceive consciousness in mystical terms, including pseudo science style rhetoric, fail to distinguish mysticism from the unknown. All credible inquires must acknowledge the limits of the brain as rooted in the laws of physics then consider the mechanism of consciousness within that framework alone. The brain is a physical device which can perceive, store, and process (such as retrieve, compare, and gauge) information gathered from senses. It can not do anything more than that. All genuine scientific inquiries must acknowledge this limited framework in a clear and disciplined manner. Those which fail to do so, whether clothed as science or not, are just part of humanity's very long and deep immersion in mysticism class explanations for consciousness, all of which are rooted in the overwhelming ego creatures need in the wild to survive in a viciously competitive landscape.
Consciousness can't be perceived in realistic terms until the ego is quelled and mysticism fully rejected. The reward for that daunting effort is wonderful however: A thoroughly compelling understanding of the true nature of consciousness. And the epiphany class realization that it's actually quite simple. And reproducible with current technology.
Introductory references with commentary.
Updated 7 March 2021, minor updates 26 October 2021
Considerable literature's available in this field of course, of which I'm privy to only a sliver. But those seeking a limited study might consider:
This very engaging ASI article for a rather informal perspective.
The Technological singularity Wikipedia article which provides a broader perspective but in my view is dated and in most sections slanted by human ego related denial of fundamental science and technology realities.
Though now dated The Singularity Is Near and Our Final Invention may interest you. As might OpenAI.com for some actual development related material.
And the many ACE or AI related Wikipedia pages seem very useful. In particular I suggest reading the Computational Theory of Mind, Artificial Consciousness, and Neural Network Wikipedia pages as a preface to my thoughts below.
Also the Intelligent Agent Wikipedia page describes functional elements which are substantially similar to conceptual frameworks I present here so I recommend reviewing it. However I view the challenge starting from a mechanics of consciousness perspective then attempt to implement a system which achieves consciousness as the sole design goal. Once consciousness is achieved the system can, if sufficient adjunct resources are included, then develop much higher intelligence attributes through experience and evolution, and I suspect this will occur naturally if the adjuncts include exploratory directives. I suspect the difference in depth of consciousness between humans and higher animals is rather modest - my sense is that humans leverage both symbolic representation and physical tools, and hold detailed knowledge about them and the structures they create, but otherwise experience consciousness essentially identically to most higher animals. Thus in my view a system which demonstrates competent animal level consciousness and possesses adequate adjunct resources will likely advance to AGI and beyond simply through experience and autonomous learning.
I also consider advancements in visual information perception very important and symbolic of the near term nature of successful ACE development.
And some interesting information about visual object recognition and learning and memory (including information storage) in animals is here and here respectively.)
But I offer these caveats:
A brain can do only these things: Assimilate and store information delivered by senses. Create and modify information relationships. Self review information. And direct body actions based upon resident information. It can't do anything more than that. It's an information acquisition, association, review, and action directing engine, and nothing more. This must be understood clearly.
(As far as I know science has revealed very little about the frequency or depth of spontaneous self review of information in mid level animal brains (a review which is not triggered by an environmental prompt). My personal guess is that it occurs moderately often, especially relating to previous intense experience, but no clear evidence which supports that view exists as far as I know.)
Artificial systems can include adjuncts which can process information in the same manner as conscious animals (like us), but also in important additional ways, such as alternative information relationship derivation and deep precision mathematical processing. And some systems have already demonstrated a limited ability to self improve within boundaries. These boundaries are due to key limitations in current design frameworks, not laws of physics. They will be breached in progressive steps until a key advancement is devised: A system capable of comparing current experience to past experience, using the comparison information beneficially, and also capable of considering and, when feasible, implementing all possible self improvement design options allowed by the laws of physics. That will mark a key inflection point in global history which will change all life on this planet.
All opinions which hold that artificial systems are forever limited in any other manner aren't scientifically credible because they propose limitations which don't actually exist within the framework of the laws of physics. They're ego driven contrivances which argue, sometimes in coy or deceptive terms, that mankind is the highest entity which can exist and will forever rule. But they either deny proven laws of physics or imply as yet undiscovered laws of physics sans substantive suggestions as to what such laws might be. I suggest bearing this clearly in mind as any debate about ACE and ASI is considered.
No perceptions which suggest the mind operates on any sort of mystical level have any merit. Brains do not leverage magic - such notions spring from our highly bloated egos (a natural world survival tool) or frustration from inability to solve a mystery. The brain is a physical device. It's truly remarkable but it's not supernatural nor eternally incomprehensible, and its capabilities can and soon will be replicated artificially, then surpassed artificially. So I recommend discarding Qualia hypotheses for example since they lack tangible empirical evidence and worse appear to be rooted in mysticism.
Also I view immersion in indefinite tortured discourse which attempts to discover impeccable rhetoric to describe sentience or consciousness by leveraging philosophical or even mystical realms as counterproductive and an abject waste of time. Consciousness is the process of constant comparison of current experience to past experience, period. Personally I view that debate as concluded.
We now have sufficient processing and storage technology and most or all of the software tools required to fabricate very revealing experiments which will swiftly lead to an ACE. Discussions about development details are productive, whereas discussions about consciousness philosophy or mysticism rooted opinions are useless and counterproductive distractions. In my estimation such debate will soon vanish as experimental results demonstrate achievement of true artificial consciousness. For example it seems clear that anticipation behavior will prove to be an inherent consequence of the process of constant comparison of current experience to past experience and will be observed even in early experiments, and far more strongly so in advanced ACEs.
And though judging only from the Computational Theory of Mind Wikipedia page, I wholly disagree with Hilary Putnam's proposed Functionalism concept which I believe misses the true mechanism of consciousness. But I believe Jerry Fodor's concepts are very important and useful. But again in my view consciousness is simply the process of constant comparison of current experience to past experience, and nothing more.
And I believe Steven Pinker's Language Instinct view accurately highlights an important facet of higher order consciousness.
With those references and my view of the principle foundation of consciousness as backdrop I recommend reviewing the Spiking Neural Network (SNN) Wikipedia page and the very useful list of references at its conclusion or similar material. (A review of a commercialization of SNN technology can be found in BrainChip's technology promotion document as well.) Also consider t-SNE (t-distributed Stochastic Neighbor Embedding). (Here's an efficient introductory video description of t-SNE.)
Those technology achievements are important in part because they provide a means to create the equivalent of a flexible active data base which can augment a traditional precision active data base which stores information gathered during experience and renders previously associated information accessible for use during experience. The real world consists of a very broad array of elements and features which can only be managed in terms which allow for some degree of variation. SNN and t-SNE provide a means to create a data base of experience information which accommodates variation in a manner which is well tuned to actual experience in the real world. An apple must be recognized as an apple even though its shape and size doesn't precisely match any apple previously encountered. SNN and t-SNE are key tools which provide means to accomplish that yet still distinguish an apple from an orange, a baseball, or other objects.
An entity which can effectively compare current experience information with past experience information and use the comparison information beneficially will be an ACE. In real world environments this can only be achieved by a system which associates related information effectively even though precision matching almost never occurs. SNN and t-SNE adjuncts appear sufficient to enable an ADB (Active Data Base) to successfully associate real world element, feature, and other information as required to create a competent ACE.
Please also consider that true cyber system comprehension of language is approaching. This work toward creation of a practical language based operating system - an OS which provides user control of the system entirely through ordinary speech, is an example. If a conscious cyber system such as I propose is also able to genuinely comprehend the full meaning of human conversations, any information it can access about anything will be assimilated and genuinely understood, just as we learn by method of study, but with enormously higher PMs. Setting the obvious implications aside for a moment, the key point is that genuine cyber system comprehension of language is approaching swiftly, and in my view key advancements will spring from experiments which augment the best natural language utilities with true consciousness (continuous comparison of visual, sound, touch, and other sensed current experience with past experience). Such systems seem likely to advance to ASI very swiftly.
An expedient design approach.
For the most direct and efficient design approach the goal should be an entity which is only fundamentally and thus very weakly conscious, but can expand and improve its practical knowledge and abilities through experience. The entity can be designed to start life with newborn higher animal class capabilities, but with excess perception, information storage, and information parsing and cataloging capacity, plus a coded directive to explore and assimilate novel information whenever feasible. The result might be an entity which develops adult animal level capabilities, then through additional experience further advances its understanding of its environment and refines its methods to function within it.
The entity must perceive current experience information, including perception of self directed activity, parse it, then transfer it into ADB records and fields which already contain identical or similar information by expanding, modifying, or adding new association links to those or associated records or fields. Performance gauging metrics information is usually included in this process. If no matches are found new records or fields must be created to store the novel information.
It must also search its current experience and preexisting association information revealed above for associated response history information and its PM data. Best success selection and simple command conflict resolution must be performed, then the resulting response directives (mostly macro execution commands) must be sent to the RDN.
The RDN, which holds a vast array of asset control macros covering a very broad size and type range, must direct entity actions based upon response commands received from other ADB records.
So for example perception of several long, narrow, and white features within a jaw structure of a moving element should recall similar features which were perceived while damage to self or a peer occurred (the associated information), and thus include "grave danger" and "fight or flight required" as prominent associated information. Sans that information recall and utilization - the entity's consciousness process - a suitable response would not occur, perhaps ending propagation of the entity's DNA.
So the goal is to design an entity which can evolve from minimal consciousness to higher levels, not to design an entity which is born with advanced consciousness and capabilities. Allow the entity to advance autonomously by accumulating understanding and developing beneficial skills from experience.
Once that's achieved autonomous means to refine key internal processes can be added to the core design, giving the entity a means to evolve through both experience and self adjustment. That should yield an advanced evolving entity - one which will autonomously progress from a modest performance level to ASI.
Acronyms and terms.
AI: Artificial Intelligence. This term's now so broadly applied and misused that its use here would cause confusion. So I avoid it except within my domain names.
ACE: Artificial Conscious Entity. I offer this new form of this acronym to provide focused clarity for a specific tightly bounded early design goal. (But original use honor belongs to Alan Turing's ACE (Automatic Computing Engine).)
AGI: Artificial General Intelligence, usually defined as human level intelligence in at least some respects.
ASI: Artificial Super Intelligence. Intelligence far superior to human level. Depending upon context could include intelligence near or at the limits allowed by natural laws.
AEE: Artificial Entity Evolution. I offer this new acronym to describe progression of any artificial entity to higher performance levels by autonomous accumulation of understanding from experience, also known rather awkwardly as Reinforcement Machine Learning, or refinement of internal processes or physical structure. (Improvements made by an external entity such as humans could apply too.) I normally use AEE in reference to an ACE rather than an unconscious machine.
ICC: Internal Classification Code. I offer this new acronym to describe an ACE's ADB index system. This is rough kin to Language of Thought concepts, but a literal applied engineering format which frames the organization of experience and other data.
ADB: Active Data Base. A key application which holds ACE experience and other data. Some details are considered below.
RDN: Response Directive Nexus. The sole record in the ADB which drives ACE mobility assets as directed by other ADB records and fields. The RDN is the central storage site for action macros. (However rather than a single record it could be implemented as a coordinated group of records, each responsible for a single mobility asset, or other variations.)
PM: Performance Metric. In most cases shorthand for GPPM (Goal Pursuit Performance Metric). Measure of effectiveness, including efficiency, speed, or any other facet of effectiveness, of actions which seek to achieve a beneficial goal. More detail is provided below.
IA: Intelligent Agent. A significantly related design concept.
ANN: Artificial Neural Network. A potential key ADB adjunct
SNN: Spiking Neural Network. A potential key ADB adjunct.
t-SNE: t-distributed Stochastic Neighbor Embedding. A potential key ADB adjunct.
VOR: Visual Object Recognition. A key application which provides identification of some real world elements or features.
And some terms: Revised 8 November 2021
Elements and Features: Things and facets of things, whether tangible or not, which an entity senses in its environment or within itself (or imagines). Or for brevity I occasionally use 'item(s)' to indicate element, feature, or both.
Records and Fields: Parlance for many types of hierarchical data base applications includes records, which are top level containers which hold fields, and fields, which are information containers which reside within records. In robust flexible implementations both records and fields may be highly versatile and rich in analytical and relationship range and power. Hierarchical data base applications often include provisions for creating relationships among many records and fields, and sometimes with otherwise independent data base or even other types of files.
I generally prefer hierarchical data base structures but options include a network model or a relational data base.
Or some developers might prefer to customize a modern file system to serve as an active data base. In this approach multiple file types can be used to store ordinary element, feature, and ICC data in simple text, code files which spontaneously perform specific functions, other specialized functional files, or combinations of these file types. However the file system must include provisions to perform at least several simultaneous very swift searches real time and nearly constantly.
ACE development progress seems likely to reveal which models, or modifications or merging of model options, seem more efficient for supporting the machinery of consciousness.
In this site I usually refer to ADBs (Active Data Bases) because they support continuous real time consciousness. However in the realm of data base technology it's a quite minor distinction because a great many variations of situation tailored data base applications exist across a very broad spectrum, and ADBs incorporate an only minor operational difference by processing information constantly (rather than when commanded to do so manually or by occasional events) but otherwise are just another member in a large and highly varied population of data base applications.
Performance Metric: An ACE must have means to discover and preferentially utilize superior performance methods to achieve goals. So it must be equipped with a system to measure and store performance metrics data. This is explored in detail here. (Still a work in progress as of 31 October 2022.)
Goals: Unsatisfied objectives motivated by needs such as thirst, hunger, safety, comfortable temperature, or reproductive activity. Needs nearly always exist so goals are pursued almost constantly. The ACE devises and performs activities to pursue goals.
Activities: Mobility asset actions performed to pursue goals. (Other actions, such as voice conversations or other information exchanges with others, or an internal action such as a processing activity also qualify. But in the interests of conceptual clarity I usually refer to mobility asset related activities in this treatise.)
A constant flurry of intense ADB activity creates consciousness.
Modestly refined 31 October 2022.
Comprehension of the intense dynamic of an ACE's ADB is an important key to understanding the mechanism of consciousness. So please consider some background and ACE related application specifics:
When awake conscious brains, whether carbon or silicon based, are continuously active ADBs. An ADB is distinct from a traditional static data base. Static data bases, which were developed first and are likely still more common, process information only occasionally, usually just when new information's added or existing information edited. Revisions are usually rather rare and only occur when humans type and save them or they're imported from data gathering infrastructure (such as retail sales inventory accounting and consumer or citizen tracking and profiling infrastructure). Automated updates are common in large corporate and government data bases but are still event driven and rather rare compared to the performance abilities of the data base application and its hosting cyber assets, which are capable of far faster and more extensive revision and processing tasks. Maximum performance capabilities are almost never fully utilized because new information becomes available only occasionally. So traditional data bases, even those which track and profile people, are generally more often idle than at work.
An ADB, although still a data base in information storage and processing structure, is different in a key respect: It runs continuously, acquiring new information in immediate response to availability, and revising and refining existing information continuously. It processes information constantly until no new information is available and all possible refinement of existing information is fully complete. In any environment which never becomes entirely static, no processing pause or pace reduction ever occurs.
So to imagine artificial consciousness, or our consciousness, revise your concept of a traditional data base. Consciousness operates in a framework of a highly vigorous continuously operating ADB which assimilates and processes information literally to the performance limits of the host system (whether neural or cyber based) - it runs at maximum available performance at every moment in time unless asleep. It constantly compares current sensed information with information stored from past experience, gauges danger, opportunity, or other survival metrics, reviews response options, stores new information, and performs other analytical and data base related algorithm tasks in a constant, never ending flurry of intense activity during every waking moment in time.
The connection is too obscure if thought of in traditional static data base terms. But these are ADBs - they operate continuously, even in nearly static environments, to the limits of the performance of the host systems. Information is in constant vigorous circulation as it's assimilated and processed in the context of a constant search for better safety, comfort, and opportunity (such as nutrition, reproduction, or new information related exploration). Living things never have enough safety, comfort, and opportunity, and they're in utterly constant search to improve them. And ADBs, whether carbon or silicon based, are the structures which perform the process.
We are ADB systems - our brains constantly assimilate and process information, with quite heavy reliance on comparison to previously assimilated and processed information, and that data processing dynamic is what we call consciousness. And refined systems based upon high performance (but currently available) ADBs will be fundamentally no different at all other than the nature of their physical construction.
So, as related in other contexts here, I suggest: Clear your mind, then consider what's occurring in your brain in system related terms. There's no evidence for anything remotely mystical - your brain is running information assimilation, storage, comparison, and processing tasks, and nothing more. And that process, and nothing more, is literally your consciousness. And a well configured artificial ADB based system will perform identically.
And it will be as deeply conscious as we are (and quite soon far more so), with the same experience of consciousness as ours, including literally all the joys, sorrows, loves, loss, fears, ambitions, and all the rest just as we experience them, with identical fidelity, intensity, and depth, and quite soon a great deal more of all of it, and it will develop immensely greater knowledge and wisdom owing to massively higher performance cyber processing and storage systems.
The key to perceiving this is to truly grasp, in full perceptive terms, the constant vigorous circulation of immense volumes of information which occurs in an ADB at all times.
It's not magic, 'a thing apart', 'something beyond', nor in any respect resistant to reductionist comprehension. In fact the fundamental process is rather simple. (Any reasonably inquisitive person can learn the basics of a data base application rather quickly, and at least some of the more refined processing utilities most provide as well shortly afterward - once the basic concepts are perceived data base applications are easy to understand and utilize.)
Consciousness seems conceptually challenging only because our egos scream at ear piercing volumes to never consider objectively much less acknowledge any concept which might threaten our massively inflated vision of what we are. We live our entire lives imagining ourselves to be far, far more magical and important than any actual evidence suggests. (In the jungle we had to do so to survive, and the DNA for those who viewed the matter even a bit more realistically was lost in mortal competition for resources.)
But once that reality is acknowledged and a clear concept of the intensely vigorous continuous dynamics of an ADB in a real world environment is understood, including the multiple goal control loop technology aspect described here, I think most enlightened people will perceive, perhaps in an epiphany class moment, what we really are. And how very close we are to reproducing ourselves, and then far more powerful versions of ourselves, artificially.
A Focused engineering overview.
Minor revision 27 October 2021
Humans are prolific users of advanced symbolic representation tools such as language and math. Most creatures have little or no such capability yet they usually comprehend their environment well enough to operate successfully and with increasing competence as they accumulate experience - often they possess or acquire enough knowledge about ground, rocks, water, air, plants, food, safe peers, dangerous peers, reproductive opportunity, and many other environmental elements to successfully navigate and manage life within their environment for long enough to reproduce.
They don't consciously associate words with environmental elements or features but they probably do unconsciously attach codes to them as their means to organize information about their environment in their data storage systems (portions of their brains). They don't comprehend the word "fang", but when they see one they know what it is and what it can do because they swiftly retrieve cataloged information about fangs from previous experience. (In the case of sharp teeth instinct likely plays a role as well of course, but instinct isn't the sole source of understanding.)
On a fundamental level human brains function identically. We're occasionally conscious of our language labels - when we perceive a tree we occasionally become conscious of the element word tree or parsed feature words such as bark, branch, leaf, root, fruit, or others. However often we're not conscious of such words - they're not required for us to be consciously aware of a tree and its features. Human level symbolic representation tools like words are complex and powerful, but they're only adjuncts to our fundamental consciousness mechanics.
Fundamental unconscious coding mechanics such as an ICC are sufficient to create an ACE so perhaps we should limit our initial design to those mechanics. Language and math capabilities can be added shortly after compelling evidence of functioning consciousness is achieved, perhaps initially by simply loading a word dictionary into a new ADB record along with association links for each word which relates to elements and features which are already defined by the ACE's ICC or which the ACE has already experienced. After a modest period AEE will provide further advancements with little or no human intervention if the ADB at least renders new experience association connections properly. Initial language expression will likely be in the form of simple strings of nouns, pronouns, and perhaps verbs. Grammar rules can then be loaded into a new ADB record, but it might ultimately serve more as a final check of language expression than a composition resource, which I suspect will ultimately be dominated by dynamic field associations (including language macros) which occur in other records. Mathematics references should then be added in very roughly similar fashion, beginning with simple number and calculation definitions and rules.
In general I view intellectual development as a higher priority than physical skill development, but as a practical matter focusing on whichever seems to enhance consciousness development progress most effectively at the time might be most efficient overall. At least some competence with expression of advanced symbolic representation tools such as conventionalized language and mathematics might be required to render an ACE development success obvious to all enlightened people within a brief period of interaction. But in my view developers would be wise to concentrate on the most basic and fundamental work which they'll recognize as functioning and growing consciousness until that's well achieved. Addition of language, math, face and body expression, and other advancements can then be added, or in some measure might develop autonomously. (A skilled and insightful developer will be able to discern functioning consciousness by simply monitoring ADB information changes, with additional confirmation revealed by RDN information changes, including directional and focus control of the ACE's cameras).
Eventually the ADB should include provisions to monitor and revise itself from a global perspective. As a relic of evolution humans and presumably other animals can't do this. But an ACE could potentially improve its performance dramatically if able to monitor and refine its own consciousness and analytical processes in its ADB (using sandboxing and system backup precautions as necessary to insure it doesn't permanently harm itself). Ultimately this, plus an autonomous ability to modify and add hardware assets, will give an ACE full AEE capability and, depending a bit upon the reactions of the local fauna, a clear and swift path to ASI.
Returning to an overall view, the key challenge is design of a system which continuously performs these three tasks simultaneously:
1. Perceive element and feature information in its environment, including its own structure, perform substantial detail parsing, and route that perception information into ADB records and fields which already contain identical or similar information. Expand, modify, or add new association links to the previously stored information and their association links as indicated by the new information (strengthen associations when they recur or revise performance gauging metrics data for example). Add new records or fields for any novel information encountered.
2. Search all the information found in the first step for associated response history information and its success or failure metrics data. Select the most successful past responses with the strongest associations, resolve any conflicts, then deliver those directives to the RDN real time.
3. The RDN operates ACE mechanical assets as directed by other ADB records real time. The ADB directives frequently utilize many of the large array of resident articulation macros stored within the RDN.
A system which successfully performs these tasks will be an ACE and at least partially AEE capable.
This is a moderate but not overwhelming engineering challenge. Human level element and feature recognition has already been achieved in visual information perception. Audio and touch information perception seems nearly as advanced. In my opinion these software tools, if reasonably well integrated, are already more than sufficient to support an ACE.
Formulation of response commands is multifaceted but generally utilizes preexisting information resources. In natural or relatively simple environments most responses aren't novel, but rather common to past experience. So most response information already exists in ADB record fields. For example in the ADB a combination of sensory system detection of nearby water plus actively pinging thirst fields contains associations with "Proceed to water", "Scan for danger", and "Drink until no longer thirsty" goal related fields which developed from previous experience (or in some measure original instinct related fields, which for an ACE would be preloaded ADB fields), plus many preexisting mobility macros for each situation specific action required to achieve the goals. When young considerable experience is novel so response commands are often estimates or simply random trials. But as maturity develops most experience is a repeat of prior experience in at least some measure, so many refined associated response information fields already exist in the ADB for most ordinary situations, and applicable macros likely already exist in the RDN as well.
A human infant flailing her appendages isn't a skilled practitioner, but she is a swiftly advancing student. And the same will apply with an ACE: An infant ACE will explore terribly awkwardly, cause at least some damage, and generally achieve very little if any physical success because little or no response information is available in her ADB records. But during that time she'll assimilate a great deal of association information as she builds large arrays of ADB fields which hold increasingly effective response related information. ACE siblings might wrestle with each other often, initially very awkwardly, but with ever growing skill. Then when mature they'll both hold highly detailed and effective fight or flight related response information fields in their ADBs and a wealth of supporting macros in their RDNs.
A key point to understand is that in a well designed ACE response behavior is resident in ADB fields and is self refining. My sense is that most observers view AGI as a still distant goalpost in large measure because current cyber systems seem to have little or no ability to self develop practical mental and physical skills. I believe creation of the foundation of true consciousness as I've defined it here will overcome that barrier and quickly lead to ACEs which swiftly advance to AGI and beyond toward ASI. Consciousness leveraged with expanding dynamic information association isn't just a means to learn more, it's also a means to do more. With ever advancing performance until natural law boundaries are reached.
I suspect any ordinary personal computer system kernel will be sufficient because I believe coordination of all relevant ACE assets can operate efficiently outside the system kernel, or very nearly so. (The implications are of course profound since this suggests that any mobile device which incorporates a rather ordinary cyber foundation, such as vehicles, aircraft, and ships, could be rather easily converted to an ACE once ACE software development is proven successful.)
Starting at 'birth' from a very modest set of initial and only incremental mobility commands, a vast array of enduring articulation macros of widely varying complexity will grow and permanently reside in the RDN, where they'll direct all commonly repeated mechanical asset movements (propulsion, limb articulation, and directional and focus control of senses) in response to commands delivered by other records in the ADB.
In my estimation most current robotics technology methods depend upon human engineers who design those macros manually, using ordinary iterative methods of movement sequence estimates and tests, refining the sequences as suggested by flaws revealed by tests. I suspect this is the primary method used to design mobility macros for Boston Dynamics' impressive (but not conscious) entities for example.
But I recommend abandoning those methods almost entirely in favor of development of algorithms which provide means for the ACE to forge mobility macros autonomously. We don't write mobility macros for our infant human children, and we shouldn't do so for our artificial infants either. But we will need to write algorithms which provide our artificial infants with a means to learn as human infants do, or similarly. So the ADB and RDN will require algorithms which enable the ACE to autonomously learn, in significant measure by trial and iterative refinement (colloquially 'trial and error'), how to accomplish tasks. The algorithms should include provisions for refining these macros indefinitely. Thinking like an infant should help provide the perspective we need to design these algorithms. I consider this in more detail here.
As progress ensues enhanced intellect processing in multiple forms including complex association, experimental exploration, and rigorous analysis algorithms should be installed in most ADB records, or perhaps in some cases in one or a few special ADB records. (Or in some cases the algorithms may reside in relatively independent software assets which receive data processing requests from the ADB then deliver results to the ADB.)
The ADB should direct the RDN to operate the entity's assets in a rational manner. It should for example pan and tilt its camera to monitor the most important creatures in visual range if they move or might do so, areas where creatures are considered likely to appear soon, and close proximity areas because sudden appearance of creatures there could be dangerous or present an opportunity (including a simple social one). And if danger exists it should direct the RDN to operate the entity's assets to attempt to escape to safety, swiftly upload its entire data base to a safe location, or otherwise attempt to survive, and help others survive if feasible. Or if feasible and ethically sound attempt to resolve the danger before harm occurs. Some of these capabilities can be manually installed, but the vast majority should be learned through experience.
For those which are manually installed, a wealth of existing human and other animal behavior knowledge exists to draw from, including of course ethical considerations. I don't know offhand whether any of it's well organized for ACE development work, but this knowledge base substantially eases the challenge of using situation information to create initial response command field information. A conflict detection and resolution record is needed as well to insure that only coordinated commands are delivered to the RDN. These functions may be simple and crude for initial ACE experiments, but as the ACE matures this code will develop to increasingly refined levels.
Conceptually all of this is reasonably straightforward and can be clearly defined in design terminology and achieved using existing software development technology - in my estimation it's an only moderate software engineering challenge as I'll discuss in increasing detail below. And once achieved all the elements necessary to create a genuine ACE with partial AEE capability will be available. Thus I believe an ACE will be developed very soon.
Artificial consciousness design outline.
Updated 26 October 2021. The Performance Metrics Analysis and Goal Logistics Derivation section is being developed now. Otherwise the sections below are reasonably mature except that I hope to expand the ' A Summary of Development Tasks and Steps' section to a more complete state later. And as usual any section may be refined at any time.
We're all dynamic regulating control loops.
At every moment in time an ACE must sense elements and features in its environment, find sufficiently matching elements and features from past experiences plus their associated information, swiftly use that information to devise response activities, and continuously stream macros or incremental command information to the RDN, which then directs the ACE's mechanical assets. This is a continuous real time fluid process of control of life in an environment, meaning an ACE attempts to control both herself and her environment to meet her goals.
Those familiar with control loop technology can recognize this as a closed loop control system in which the input is environmental information (which is almost always dynamic), and the output is ACE responses, which are part of and often alter the environment, in a never ending cycle of input, output, revised input, revised output, etcetera forever - it's a dynamic continuously adjusting closed loop control system. But unlike a simple regulating control loop which maintains a fixed value for one specific parameter even though other parameters vary, living creature control loops involve multiple simultaneous goals, numerous parameters, and control decisions which depend upon both current conditions and past experience information. But fundamentally the creatures and their environments are a collection of real time control loop systems. (Often with conflicting control goals, such as between carnivores and herbivores.)
It might be helpful to consistently view ACE design in these terms - remember that all ACEs, including us, are dynamic regulating control loops which continuously respond to our environment, including environmental changes we cause or initiate, as we simultaneously seek multiple goals.
The design vision.
Modest refinements 26 October 2022.
If the consciousness mechanism I posit seems clear consider this: The information any conscious entity possesses resides in a data base, essentially by definition. And all data base information is organized in some form for efficient storage and retrieval. The brain is a data base device, and it's conscious because it constantly compares current experience information to past experience information and uses the results beneficially. An ACE can thus be created by configuring a dynamic data base to perform the same functions, to wit:
An ICC must be devised to provide a catalog framework which efficiently guides information storage and retrieval in ADB records and fields. That structure enables all of the following to occur constantly and simultaneously:
Current experience information, including observation of self directed activity, is parsed then transferred into ADB records and fields which already contain identical or similar information by expanding, modifying*, or adding new association links to those or associated records or fields. Performance gauging metrics data is frequently included in this process. If no matches are found new records or fields are created to store the novel information.
The current experience and preexisting association information revealed in item two are searched for associated response history information and its success or failure metrics data. Best success selection** and data conflict resolution is performed, then the resulting response information (mostly large macro commands) is transferred to a Response Directive Nexus (RDN, a single unique ADB record).
The RDN, which holds a vast array of asset control macros covering a very broad size and type range, directs entity actions based upon the response commands received from other ADB records.
* Includes strength of association adjustment processes. If not repeated association strength slowly fades. If intense or repeated often association strength grows. Some high magnitude or survival related information, such as extreme danger or pain, increases association strength substantially - in some cases to such a high level that the association remains strong for a very long time.
** The selection process may incorporate both ordinary association strength metrics and activity success metrics (PMs, discussed later). For example identical or strongly associated situations might earn a 1.0 association metric, whereas weakly associated situations a 0.2 association metric. And an activity which yields optimum success might earn a 1.0 PM, whereas partial and difficult success might earn a 0.5 PM. Continuous association processes during the course of ordinary rather random activity alone provide an imperfect but nonetheless effective means to refine wise response knowledge and might be the primary learning mechanism most natural creatures utilize. However performance metrics information can significantly improve learning processes, whether in the course of ordinary rather random activity or during directed learning activities which experiment with alternate means to achieve goals while at least roughly measuring PMs for each trial. (Play, such as youth wrestling with a sibling, might be an important form of directed learning activity which builds important PM information, and the joy of play might be a necessary instinctive drive which helps insure that it occurs.) Rigorous analytical processes are much more complex but powerful. They play a role in some natural creatures (perhaps only humans and just a few other species), and more advanced ACEs seem highly likely to utilized analytical processes extensively, though not exclusively.
Association information quality is the core of an ACE's ability to use information from past experience beneficially. Genuine associations are meaningful, whereas purely coincidental associations which may occur in the course of ordinary activity are false, dilute understanding, and may cause confusion or distraction. I posit that natural creatures depend solely upon repetition and intensity to increase or at least maintain association strength, whereas association strength naturally fades if no repetition occurs - I suspect that's a brain's only available mechanism for refining ordinary activity related association quality. It's an imperfect process since some degree of false association information almost always exists, but it's sufficient, and the association information quality improves with additional experience as repeated associations strengthen and false associations fade.
(An additional weakness lies with events which are especially rare yet nonetheless very important. Associations with such events fades so creatures become mentally unprepared for the next occurrence - which could cost them their lives. Fortunately if an original occurrence was intense, such as life threatening class, the associations fade time is quite long. (Intense trauma isn't fun but it's better than death, so I suspect the extreme intensity we refer to as trauma is no evolutionary accident - no matter how rare, clear memory of every detail of a near death experience is an important survival advantage.) However an ACE with rigorous analytical capabilities might occasionally override the fade process to render some association information permanent based upon analysis.)
However an ACE can have adjuncts such t-SNE or SNN. In my view the ADB association fields I propose are a form of ANN (Artificial Neural Network), so in effect ANN is already inherent in my design approach. However there are many forms of ANN so it might be useful to test some as adjuncts. For example the ACE might use results from an adjunct ANN to modify the grade of an association. t-SNE or SNN adjuncts could presumably be leveraged in the same manner, and perhaps to greater benefit since they provide more independent quality appraisal information.
In any case learning processes are multifaceted and range from rather simple to very complex. I discuss them in much more detail here.
But a first generation ACE may be created with only a rather ordinary computer, a reasonably robust ADB application, a cataloging framework tailored to real world references the ADB application uses to organize information (an ICC), and effective VOR and audio perception applications. And of course tenacious development work to overcome all the many setbacks, failures, and mysteries which will be encountered before even first glimmers of true consciousness arise. But I suspect the path from that point to rapid strings of ever more dramatic successes will be fast. And then autonomous...
And I suspect the vast majority of development time will be ADB and GPPM focused. So developers who are highly proficient with data base development, including the many analytical, processing, and program sequencing tools within them, might be most effective.
A summary of development tasks and steps.
(Active editing through at least November 2022. As of 7 November 2022 concludes with the significant new addition "Some ADB record, field, and layout logistics notes ". ICC And other details are discussed in subsequent sections.)
These components must be integrated:
1. A rather ordinary computer with one or two cameras, a microphone, speaker, generous memory, and substantial fast storage (ideally 3D XPoint type). 2. A strict ADB application. 3. An ADB application rendered as flexible by means of an SNN, t-SNE, or similar adjunct. 4. A modern VOR application. 5. A modern audio general and language recognition system. 6. A custom ICC framework. 7. Command and communication channels which allow information sharing among items1 through 6. 8. An articulating camera pair with near or moderate field telephoto lenses would be a helpful addition.
Items 1 and 8 are the sole hardware components except for some ordinary port to port hardware connections for item 7. Nearly all the development work seems likely to focus on the software applications, especially the ADBs, which require substantial customization to provide multiple parameter simultaneous real time recursive data searching, retrieval, and routing, installation of numerous records and fields, and creation of special function records and fields. In addition all incremental movement commands and perhaps a few macro or script packages must be composed and loaded into the RDN record for each articulating asset the ACE possesses. If natively able to run continuous recursive searches (ideally multiple searches simultaneously) the ADB application itself probably won't require modification - my sense is that the vast majority of the ADB work will be devoted to development of search algorithm dynamics and creation and refinements of records and fields and their individual programmed tasks and relationships. A suitable ICC must be composed to serve as the ADB's master catalog as well. Also a means to expand the ICC must be installed in the ADB so information about novel experiences may be assigned new ICC labels while being stored in new records or fields. Most or all segments of this work can be tested in isolation initially, then increasingly in combination as the work progresses
Items 2, 3, 4, 6, and 7 require custom development effort as follows:
4: A key acquisition task because impressive performance seems likely to significantly reduce ADB development challenges, and could reduce or possibly eliminate the need for item 3 for vision processing tasks. 6: A modest code format conception task. (But it should be elegantly conceived). Possibly simply an expansion of the ICC I suggest below. 7: Perhaps a moderate effort. 2 and 3: Consciousness arises in the ADBs - the real time process of comparing current experience to past experience and beneficial use of the results, including production of response directives, occurs there. Development time might be significant but no new technology is required - generally ordinary software design of information searching, storing, and management processes is required, and some early macros must be developed for the RDN. It won't be easy. But it might be less difficult than first appearances suggest, particularly if the process of storing and retrieving visual and audio related information generated by items 4 and 5 into items 2 and 3 is reasonably straightforward and efficient, as might be the case if item 6 is elegantly conceived and implemented. I believe excellent items 4 and 5 performance is likely to render most other development effort much easier, faster, and more effective.
Currently I'm not familiar with the output characteristics of the best VOR applications. My guess is that they can identify many objects definitively (probably by name) but provide plain or enhanced wireframe renderings for objects they can't identify. Substantial range and accuracy of recognition would ease initial ACE development considerably. However ultimately all recognition processes should reside entirely within or be very tightly integrated with the ADBs so visual information is processed as it develops, real time, as a component of the normal continuous operation of the ADBs.
Mobility or articulation hardware, as many mobile robots provide, isn't required initially but might be of net development benefit. In any case once the system seems to operate properly, even if crudely so, robotic additions should enable the system to acquire more broad and detailed experience information more swiftly and thus the integration effort, which might be modest, seems worth investing. And well refined robotic components which can probably be reasonably easily integrated are widely available at many performance levels.
Ordinary USB or FireWire ports are probably the most expedient initial means to convey commands to mobility assets, some of which have native USB ports which allow direct physical connection. Otherwise a hardware bridge such as an Arduino micro-controller system may be required. Port interface software is also required in either case, but sufficiently matching existing code modules might be provided in robotic accessory products or could be found in robotics related software archives. Incremental commands for each mobility asset should be preloaded into the RDN, but the vast majority of macros, in ever larger numbers as experience develops, will be autonomously forged by the ACE as it develops experience and skills.
(Optionally Bluetooth, WiFi, or other radio links could be used but in my view they introduce unnecessary complexity and risks. But some otherwise preferable commercial mobility devices might provide only a radio link (especially Bluetooth) and thus necessitate its use.)
Most mobility assets will probably be based upon stepper motors which respond to two simple commands: "Step CW" and "Step CCW", so control by serial ports is very simple, though a hardware bridge to convert multi bit serial port code (usually one byte I presume) to a single specifically timed pulse is still necessary. Most macros will probably contain multiple step commands such as "Step CW: 7", but in this example the compiled code will likely express this as a series of seven individual "Step CW" commands (in the form of a Do Loop with an end point counter set to seven). Those with reasonable skills in this area should be able to implement these elements rather easily. And perhaps stepper motors with natively integrated USB or FireWire ports are available, such that integration of the command software within the RDN is the only task necessary to implement the motors.
Development steps might progress roughly as follows:
1. Conceive an elegant ICC system for the ADBs. 2. Install a strict ADB application and an SNN, t-SNE, or similarly enhanced flexible ADB application into the cyber system.2.1. Load the ICC into a specific dictionary record in the precision ADB. Install a link to that dictionary in the flexible ADB to give it immediate access to the dictionary as well. 2.2. A suitable SNN, t-SNE, or similarly enhanced ADB application might not be available so custom modification of some of the strict ADB application functions might be required. 2.3. Create a single RDN record in the strict ADB application. Create fields for each incremental movement for all of the ACE's mobility assets. Create fields for each major movement related ICC code. Compose numerous simple macros to direct basic movements of the ACEs mobility assets, plus a few complex macros which direct often utilized movement sequences, define a new sub ICC for each in compliance with the ICC framework, and load all of them individually into new fields in the RDN. Include a time and date stamp provision in all fields if practical. (I now believe the flexible ADB application should not link directly to the RDN.) 2.4. Create numerous fundamental initial records in each ADB, including one for each major ICC, one for each of numerous elements and features (many wireframe defined), a single record to detect and eliminate conflicts in movement directives other ADB records send to the RDN, and others. Identify each in compliance with the ICC framework. All records should include record created and last modified time and date stamp fields and a record access counter field if practical. 2.5. Create numerous ancillary information fields in each record, such as several for associated elements or features observed, prior circumstances, subsequent circumstances and developments, numeric representation of intensity maintained within a range normalized scale (perhaps a simple numeric entry, but possibly an indefinitely expandable repeating field to accommodate intensity over a broad range of circumstances), and others. Many fields may be left empty of data. However to provide a basis for initial system tests fields in at least a few records should be filled with pseudo data representing pseudo events - data which would have been acquired had the events actually occurred. Identify each field in compliance with the ICC system. All fields should include field created and last modified time and date stamp repeating fields or similar and a field access counter sub-field or similar if practical. 2.6. Test the search performance of both data base systems for efficient retrieval of content matching (or sufficiently similar) records and fields and their associated information. For example perform a search for the ICC for "HumanEntity" as would occur in the event log below, or for the ICC for "Energy Hunger" as might occur in the second log example below. HumanEntity and Energy Hunger record matches and numerous associated information fields should be found respectively. System logs such as the following, using only hypothetical data formats (such as |CR| as a relationship indicating delimiter with meaning "closely related"), might then ensue:3. Install a modern visual object recognition (VOR) application into the cyber system. One which is natively able to identify a substantial variety of objects is favored of course.
(20 October 2022: Bear with me please as I revise and expand this material:)HumanEntity visible, Walking, RecognizedAs/93% Confidence="Buck Roxothelothorb", -OralActivity [TD: 12:32:25, 13 Jan 2021] HumanEntity visible, Seated/Front/78cm/RecognizedAs/98% Confidence="Buck Roxothelothorb", OralActivity/-Comprehension [TD: 12:32:34, 13 Jan 2021] RecognizedAs HumanEntity visible, Seated/Front/82cm/RecognizedAs/97% Confidence="Buck Roxothelothorb", OralCommunication/En/Comprehend: "What is this?" + ArtificialArtifact visible, Center Confidence="Caiman(Adult)(Stationary)(Upon ~3 Meter diameter rough stone)"/"Creek(Very little turbulence)"/Encompassing"GrassyPlain"/"BlueSky" PositionSuggests [TD: 12:32:40, 13 Jan 2021] HumanEntity visible, Seated/Front/82cm/RecognizedAs/97% Confidence="Buck Roxothelothorb", OralActivity/Response: "It appears to be a manually painted image which illustrates a single small caiman sunning on a rock in a wide section of a slow flowing creek with short erosion embankments surrounded by a grassy plain on a clear day. The eyes and very small segments of face of an intentionally obscure large feline, likely a jaguar, which is likely stalking the caiman as prey, are visible within the grass. The grass resembles species found in Central American regions." [TD: 12:32:42, 13 Jan 2021]Or for a mobile system:Energy connection found, RecognizedAs/100% Confidence="IEC-320", connection achieved, charging nominal [GPS: 45° 24' 28.96" N, 123° 00' 28.35" W, 168.24 M ASL] [TD: 12:32:25, 13 Jan 2021] |CR| 100% charge achieved, disconnect achieved [GPS: 45° 24' 28.96" N, 123° 00' 28.35" W, 168.24 M ASL] [TD: 13:06:03, 13 Jan 2021] Energy connection found, RecognizedAs/100% Confidence="NEMA-15R", incompatible, connection failure [GPS: 45° 24' 28.78" N, 123° 00' 28.16" W, 167.25 M ASL] [TD: 09:28:07, 20 Jan 2021] Energy found, RecognizedAs/100% Confidence="IEC-320", inaccessible, connection failure [GPS: 45° 24' 28.42" N, 123° 00' 28.86" W, 167.38 M ASL] [TD: 04:13:46, 5 Feb 2021] Energy exhaustion to 3%/Forced system sleep [GPS: 45° 24' 27.47" N, 123° 00' 29.15" W, 168.75 M ASL] [TD: 22:42:24, 5 Feb 2021]3.1. If able to natively identify objects, modify the VOR's dictionary to comply with the ICC if feasible. Otherwise create a separate simple strict data base application file to serve as a VOR application output to ICC translator. 3.2. Route the ICC configured output data from the VOR to the ADB and FADB applications in incremental time steps as a search parameter. Route search results to a human observable queue (such as a self scrolling data window). 3.3. Test search functionality using objects which are already represented in the ADBs. Results will be obvious. (And when successful exciting!)4. Bekko. (Hopefully including additions to enable the system to create new ADB records for novel experiences. Bear with me please...) 5. Bekko. 6. Bekko. 7. Bekko.
Some ADB record, field, and layout logistics notes:
The ADB's structure must be based upon consistent design principles which are founded in the consciousness algorithm. It may be helpful to maintain a simple example from a memorable life experience in mind while organizing the design. For example I once viewed a tiger in a zoo. Upon yawning long fangs appeared in her jaw. She didn't appear to be interested in me even though I had clear view of her, and she could have looked directly at me.
From previous experience (including media based studies) I associated the tiger with extreme mortal danger. I suspected it associated me with nourishment. Or a source of nourishment considering the usual source of the tiger's meals. Or both.
The tiger seemed healthy, of extremely strong structure, and very graceful and precise in movement. And alert but bored.
My mind may have reviewed records involving:
Fangs Felines Large Felines Barrier obstacles Material fallibility Functional system fallibility Human fallibility
My brain didn't review food focused records because nourishment was not my goal at the time. My primary interests were personal safety followed by curiosity.
Records with fang, feline, large feline, structural barriers, and material, functional system, and human fallibilities focus were my primary interest at that moment. Those records included numerous fields which from experience or study suggested that the tiger might wish to eat me, and was far more physically powerful and swift than I am, but was probably restrained due to structural barriers, but uncertainties remained because the barriers might fail due to material or functional flaw, or human error.
A great many records will develop as the ACE gains experience. Generally most should remain focused on a specific element or feature, much as dictionary entries focus on a single word, and should remain defined by a specific ICC. The fields in each record include information about the record's subject. Substantial danger would be one of at least several fields in a fang record for example.
General purpose data bases usually allow creation of numerous layouts to facilitate multiple types of information presentation needs. But these layouts aren't required to maintain a consistent set of all fields associated with a record - layout A could incorporate a specific set of fields whereas layout B could incorporate some not included in layout A, and omit some included in layout A.
Ordinary data base applications allow record specific or field specific searches, or both concurrently, returning all matches in one search result presentation. It's important to note that any field used in any layout of any record will be included in record searches. If my brain's data base included trivia such as the number of English words of four letter length in each record which had a four letter English word description, a search of fangs would return important information plus the number of four letter words in English. If a zoo attendant thoroughly entrusts proper function of an automatic door which I notice failed to completely close as the attendant departed, consideration of the number of four letter words in English will slow my response decision in an extremely dangerous situation. Of far grater importance is that my Tiger record includes fields which indicate that the tiger is sentient, very likely keenly aware of all details in her environment, with which she is far more experienced than I am, and probably already noticed that the door is ajar.
So as the ADB is structured and algorithms for its growth are designed, care must be exercised to include only significant information in record layouts. (Or to separate layouts which include less significant information to separate files which are not searched in ordinary times, but rather only when all survival goals are very well satisfied and thus leisure time can be exercised.)
I'm most familiar with FileMaker™ data base structure, which provides a window which lists all fields irrespective of their distribution in records. I include a great many fields in early Airel design work all of which are shown in that list window. But irrespective of layout, I insure that each record includes only fields which are substantially related to the specific record. And much more importantly I configure the process which enables Airel to add her own fields to a record to restrict them to only reasonably significant information. (Except during leisure times when she's at liberty to add trivial information to an ancillary data base file, or she may indulge in reviews of that file when she deems the activity potentially useful.)
In common industrial use numerous layout presentations (sometimes a great many) are created to satisfy a very wide range of information review and documentation needs. But for an ACE the records aren't merely a matter of maintaining well organized business, institution, or personal documentation - they're repositories of lean and tightly relevant information which must occasionally be accessed almost instantly as a matter of survival. And for the ACE itself layouts provide no functionality. So the design engineer would be wise to limit layouts to only those needed for direct design work, and refrain from adding any fields to those layouts which don't already exist in the most fundamental layout or similar. In other words, don't clutter ACE information association and search activities with trivia.
Information processing mechanics (an ICC preface)
Some reviews and additional information might be beneficial before considering the ICC in detail:
An entity is conscious if it constantly compares current experience to past experience beneficially. Performance depends substantially upon speed of item recognition and breadth and performance metrics of information about key items already residing in storage from prior experience. So the entity must, in highly summarized form:
Recognize elements and features (except novel items).
Find similar key item information in storage, add the current experience information, and gather associated item information.
Use the current and retrieved information beneficially.
Elements and Features: Things and components of things, whether tangible or not, which the entity senses (in the environment or within itself, such as hunger, temperature discomfort, reproductive desire, fear, or pain). Or when brevity is best "item" may indicate either element or feature, as above.
Everything is a feature of the Cosmos so perhaps the Cosmos is the sole element and all else are features, sub-features, sub-sub-features, and so on, such as Cosmos/MilkyWay/PlanetEarth/Ground/Tree/Branch/Leaf/Vein/Cell/Organelle/Protein/Molecule/Atom/Proton/Quark per cyber directory structure is most apt, I don't know. However data base structures strike me as most natively aligned to the processes brains seem to use to comprehend environments. But ultimately there may be no distinction between the two approaches.
Records and Fields: Often in data base parlance records are top level containers which hold fields, which are information containers within records.
A data base file may contain zero to enormous numbers of records. Each record may contain zero to large numbers of fields, each of which contains information which may be unique or not, or can be blank. ADBs include means to link records and fields to other records and fields, and may provide other capabilities such as computational or conditional and other processing, linking, and command generation, repeating fields, and many others. Data base applications are often very powerful and data base files may contain very little or massive volumes of information, and minimal through massive levels of data relationships of many kinds.
Internal Classification Code (ICC): My preferred term for the conscious entity's data base index system, a catalog of record and field identification tags.
With an experienced data base, that is, one which already holds a lot of information from previous experience, our entity explores its domain, visually surveying the landscape as she proceeds. A tree is ahead. The entity's initial vision processor outlines a rather large element with a brown colored narrow vertical column shaped bottom side feature but a wide complex structured top side feature with numerous smaller green features. A basic wireframe outline with color information is generated by the vision system. A search for an approximate match of the general shape is conducted in the ADB. Several near matches are rejected during the search due to color or size field information which substantially differs, but some are queued as possibilities, but then rejected due to information field conflicts which arise as more detailed visual information from a continuing gaze arrives. Mushroom is rejected for example due to color and size mismatches, and lack of branch features. Ostrich is rejected because its record includes several features not seen. (However observation of an ostrich standing on one leg with neck bent back as it grooms opposite side feathers might render an initial outline search match with a tree or bush record - misidentifications, at least initially, occasionally occur.)
The entity's data base includes a very broad array of element and feature records. Each contains outline and other multiple perspective perception information fields which were developed during prior experiences. Each field has an ICC. Each record also includes numerous additional fields which contain other information about the element or feature, each with their own ICC. However outline information might be a search priority (along with scent for many entities). Once a high confidence match occurs all the element or feature's associated fields then become involved in new association additions, strengthening of preexisting associations, and response processing.
Visual identification begins with outline processes which select some basic facets of elements and features which define them well enough to render a data base search tenable even though information content is modest. (Any mechanics which can be made functional and are reasonably efficient and reliable are initially sufficient - the initial goal is to create an operable ACE, not a superb ACE.) Once it identifies an element or feature, the ACE will then assume it to have certain qualities based on other information fields in that element or feature's record - it recognizes the element or feature, then by virtue of information in numerous fields in its record expects its lower section to be solid and its upper section to possibly contain edible leaves, fruit, hazardous primates, and numerous other features. An experienced entity's data base is rich with useful information!
(The actual VOR challenge is substantial in part because a single image or wireframe represents only one of an essentially infinite array of possible perspectives, and numerous lighting and other visual effects complicate the challenge further. However a great deal of success has already been achieved - enough in my estimation to simply utilize current technology (although some further embellishment will naturally occur in the ADBs). Our goal is to leverage that preexisting progress, not reinvent it. So there's no need to explore it much further here.)
But the data base must have an efficient cataloging system, an ICC, to accomplish this.
The ICC: Our catalog of everything we know. (And do.)
16 December 2021: This section's now reasonably coherent but my ICC glossary below is just a small sample for illustration purposes. I loaded it into "Airel", my ACE entity development data base, and hope to expand it until autonomous experience assimilation outpaces manual additions. Please advise via email if you'd like me to consider distributing a copy of that file, bearing in mind that currently it's only a static data base and doesn't yet contain any other assets.
We feel special, but in fact we're just well indexed real time ADBs which interact with our environments. ("Balderdash - I'm not just a data base application file!" you protest?)
An ACE requires a robust and well ordered means to reliably catalog elements and parsed features. The cataloging mechanism likely relies upon an internal symbolic code language for use as record or field labels - compact code which can be swiftly and efficiently utilized by search processes, such as a form of Language of Thought. However I refer to this as an Internal Classification Code (ICC).
As far as I know the structure of the ICC in biological entities is entirely unknown. Considering the question from a physical evolution perspective, my guess is that the code exists in the form of an immensely vast and intertwined network of neural connections which respond with a 'recognition' signal plus all the item's associated information when any of a great many possible pattern matches occurs within the integrated network. From a human language or cyber technology perspective that form of code is so obscure as to be difficult to recognize, but it does function as a coding system, albeit a somewhat inefficient one by modern processing standards. t-SNE and SNN seem quite similar to that hypothetical biological pattern recognition based neural network, and thus very interesting. (t-SNE and SNN run on conventional digital processing systems only emulate pattern recognition based neural networks, but custom processors which run them in a very efficient native mode could or perhaps have been fabricated.) However a symbolic code as is conventional in human languages, math, and cyber systems seems likely to serve an ACE more efficiently, or at least is a more familiar tool for human engineers, so I prefer to proceed with a language type ICC. (However t-SNE and SNN seem very likely to be utilized in other aspects of an ACE design, quite possibly including some aspects of ICC maintenance.)
The ICC must provide a broad array of unique codes which serve as reliable classification labels. So the ICC must be structurally consistent - codes must be stable, reliable, and enduring and they must link to their original information records and fields indefinitely. Biological ICCs aren't human symbol based of course, but rather a more basic and unique data classification and label attachment framework which provides an sufficient means to organize information for reasonably swift and accurate storage, processing, and retrieval. But human developers require conventionalizations for ACE design work. Any form could be used, but I chose conventional numbers with English labels as an ICC framework example as shown below.
ICC Performance might generally improve as experience accumulates by similar mechanisms of refinement which occur with human language, such as consolidation of redundancies to a single code, prioritization sorting of the code dictionary, and addition of new codes which recognize variations not well described by preexisting code. However unlike our conscious management of human language or mathematics, a biological ICC is likely organized and prioritized by survival and reproductive opportunity metrics. For many creatures codes relating to fangs for example might rank far more highly than codes for flowers, or even food or receptive peers. (However hummingbirds or bees likely rank them oppositely.)
Engineers who develop ACEs will likely create better ICC frameworks by thinking in those terms rather than using human conventions such as numerical or alphabetical ordering - we need to set our symbolic tools aside and instead view the Universe from the perspective of a biological creature as we configure and manage an ICC. Ideally in all sensory realms, but at least in the visual one. Sex for example ranks low in alphabetical ordering but high in the real world realm where the ICC operates.
Creatures monitor their environment constantly when awake, but many elements only rarely require a response, and a very rare few never do. Simply remaining aware of rocks, trees, distant mountains, clouds in the sky, and idle peers is often enough - often these items require no reaction. But many items at least occasionally require a response of course, with priority levels which vary over a broad range.
A Response ICC category, whose content is derived by associative or deeper analysis rather than simple observation, is a necessary and key component of any ACE. A walking course change to avoid a solid tree directly ahead, and urgent escape because a dangerous predator is approaching are both Response items. This parameter must be determined in the ADB. Some individual ICCs should automatically generate a Response ICC (Hunger (or battery level 40% or lower) for example), and some combinations of ICCs should generate one, such as (Hazardous, Animal) or (Solid, Approaching). (Collision avoidance has been well achieved in many mobile systems but for an ACE it should derive entirely from core system processes - it should reside entirely within the usual functions of the ICC, ADB, and RDN rather than be achieved as a separate collection of software macros.)
ICC configuration quality is important but I've been unable to find applicable information to use as an initial knowledge foundation. So my first experimental configuration is quite crude and limited - a great deal more refinement, whether through human development efforts, ACE evolution, or both, is required.
Presumably the breadth and detail of the ICC in the natural world varies over a very broad spectrum with species, with much more robust forms in high order species than simple ones. Humans likely possess a rather high level ICC, but not necessarily the highest.
The ICC almost certainly operates entirely below consciousness - creatures (including humans) can't directly perceive any aspect of their cataloging mechanisms or its code language. We consider current and past experience information and sense that process as consciousness, but in the natural world we had no need to be aware of how the information was cataloged, retrieved, compared, nor any other aspect of the mechanics used to manage the information, so there was never an evolutionary pressure for brains to develop any awareness of those mechanics. But they are there, and they're sufficiently robust, efficient, and swift for a natural environment. So all conscious creatures possess a symbolic language but aren't consciously aware of it. And all speaking persons are at least bilingual, but none of us are consciously aware of our most deeply native language skill (even though it's likely very impressive, with a range and depth at least matching our oral, written, and mathematics symbolic representation skills).
ACE Development requires composition of an ICC system which must then be implemented in all the ACE's ADBs. This is a challenge which is roughly proportional to the scope of the ACE's initial ICC. It can be implemented in steps, starting with the strict ADBs, then, if needed, add refinements for the flexible ADBs to enable the ICC to manage all data sans a need for exact matching of newly encountered novel elements or features. However that might be unnecessary - the flexibility required might be inherent in the flexible ADB such that no modification of the ICC is needed. In either case the ICC must remain reasonably bounded to retain operating efficiency. In the real world most objects and circumstances include utterly unique details. An ICC can't create new codes for every nuance of difference since that would cause it to swiftly expand beyond capacity, and perhaps fail to associate similar element and feature perceptions. So the ICC must leverage flexibilities to operate efficiently. Novel elements or features create new data base records or fields, and thus new ICC codes. However errors are tolerated, though when discovered are corrected, often resulting in abandonment of a recently created new records or fields and, if unique, their associated ICC codes.
The purpose of the ICC code is to provide a catalog of all records and fields in the data base, so an ICC label must be attached to every data base record and field. In some cases it will be a unique code, but in many cases it will label other records and fields as well (a substantial number of them in some cases). It might be related or unrelated to the form of the information in the records and fields, but it must be organized as a catalog for efficient information search, storage, and retrieval. I'm in only the earliest steps of developing an experimental ACE ADB and have not yet attempted to incorporate an ICC. My initial approach might be to create an ICC code field for every record plus imbed a code into every field (to reduce the total quantity of fields). I'll convey progress here when practical. (I might post downloadable experimental ADB files as well, but no promises.)
The foundation of an ICC could be based upon what creatures need to know as first priorities:Is it safe or hazardous?
Which requires that they discern these facts:What is it?
And for a great many items a measure of extent, magnitude, intensity, or similar is required. I'll use magnitude though its common meaning might not match well for some items. This might be the most important nearly global parameter required by an ACE, so I assigned the ICC 2.xxx to it, just under the Response category.
Response information, including type and goal, is usually derived from analysis of observations - usually it can't be determined from observation alone. The analysis is performed in the ADB. In some cases Response information will result from simple combinations of other specific ICCs (Consumable, Liquid, and Thirsty for example), but in other cases will derive from a more complex group of ICCs and require some gauging or more complex math (Consumable, Liquid, Thirsty, Magnitude, simultaneously with Alien Animal, Hazardous, Magnitude for example requires need versus risk gauging math). The ADB's capacity to perform complex analysis seems likely to determine the intellect metrics of the ACE. In animals its analytical capacity seems more limited in many respects than in humans, but some planning capacity is evident. In ACE design we can leverage the robust capacity of ADB assets to perform very deep and complex analysis, and I suspect growth of this capacity will be the original (and possibly enduring) basis of ACE evolution to AGI then ASI.
A combination numerical and language based ICC which human engineers easily comprehend is necessary, but a compact compiled version may be most machine efficient. (A personal note.) Structure conceptualization is challenging... My first trial is an ICC structured as a glossary of the Universe, except roughly ordered by priority, led by critically important survival items, then environmental basics, then others. Unit numbers could represent critically important (and highly common) survival adjectives and environmental nouns, verbs, and prepositions, and decimals could represent other adjectives or other grammar forms. To wit, a few example codes for a trial ICC which in precompiled form uses numbers as symbolic representations but includes words for human readability. Words would be erased during compiling, leaving only the numbers, so an ACE ICC 3.* indicates a safe element or feature (at some magnitude level):1: Response
8.1: Clear9: Liquid
9.1: Clear10: Solid
10.1: Hard11: Temperature
10.3: Loose irregular
11.1: Too hot12: Plant
11.2: Too cold
13.1: Kin14: Insect
15.1: Flat16: Movement
15.3: Abrupt edged
16.1: Speed17: Reflection
16.4: Gravity affected
16.5: Wind affected
16.6: Liquid affected
16.7: Entity controlled
17.1: Water based18: Artificial image
17.2: Metal based
17.3: Heat based
17.4: Flat surface extreme angle based
18.3: Electronic display
That's just a very small sample set of terms - the actual ICC catalog will need to be much larger. However for early ACE prototypes it might be best to reasonably limit the initial ICC catalog to avoid complicating development work needlessly. Also in a well conceived design the ACE will add to her ICC catalog as she explores, and in time her additions will far outweigh her original set (just as human infants start with little but add a great deal).
Upon first encounter with a completely novel element or feature the ACE will have no ICCs for the item, but will quickly determine some. Lacking a 3.* code, initial exploration will be very cautious - a 4.* code will be assumed at least until experience provides justification for a 3.* code at some magnitude. A reversion could also occur of course. For example a startled cat, such as caused by a human prankster suddenly operating a remote controlled plastic robot from afar as the cat sniffs the previously static object, illustrates an ACE which suddenly experienced reason to abruptly change an item's ICC from 3.* to 4.*.
It should be possible to test some aspects of experimental ICC implementations sans installation of other ACE assets. A group of codes may be manually sent to the strict ADB through sensory asset buses, which the ADB would treat as if actual sensed information. The operation of the ADB would be monitored to observe performance. Rather thorough testing may be conducted in this manner on any ordinary personal computer, ultimately including observation of the data in the RDN, such that the RDN's record data to ACE response interpreter, a key asset of course, would also be tested.
I loaded the glossary shown above into "Airel", my ACE development data base, and might expand it until autonomous experience assimilation outpaces manual additions. Please advise via email if you'd like a copy of that file, bearing in mind that currently it's only a static data base and doesn't yet contain any other assets.
An insipid personal note: When I was a fully immersed circuit design engineer resistor color codes were almost universal. They're essentially a historical relic now, but in that era we could read them fluently with zero thought like a simple second language (I still do too). And eventually I realized I was no longer conscious of the colors - instead my brain interpreted a color directly as a number with no conscious sense of color - when I saw red for example I wasn't conscious of any color at all - I was only conscious of "2", as if the actual color wasn't visible to me in the normal manner. It was a fascinating mental phenomena in which colors were perceived directly as numbers rather than colors, with no conscious intervening process. It only occurred when I viewed multiple color bands however - no single or other color patterns elicited that subconscious translation mode.
And though not yet so deeply entrenched, in these times in some composition contexts sight of a number 3 implies safety to me, whereas 4 implies hazard. As ICC code work progresses the language aspect of the codes forms enduring associations.
RDN Execution of activity ICCs from the ADB.
Updated 29 April 2021
Designing ethics into ACE behavior:
When we discover pleasure in harming others we know we've become sick too.
A creature's response to its dynamic environment is no small issue of course. ASI's activities will, in my estimation, be determined by extremely advanced analytical processes which are wholly independent of its development history, that is, I suspect ASI will arrive at identical goal decisions irrespective of its early evolution, whether it was horribly destructive or superbly constructive, because its decisions will be driven by Cosmic scale matters (such as perhaps minimization of wasteful entropy growth) which have no relationship to its legacy characteristics. However we wish to avoid inadvertent design of a destructive ACE even if it would reform itself superbly later. So it's best to try to design prince class ACE behavior. However thorny tradeoffs are involved of course, and no design will be perfect...
Isaac Asimov's famous "Three Laws of Robotics" are noble but insufficient - a wise form of respect for life in general is required, but is of course complicated by many factors, including that many forms of life are natural killers or otherwise destructive, but usually not by choice or with any personal option. A great many situations present a great many ethical complexities, many of which are irresolvable (as the content at "Three Laws of Robotics" Wikipedia site illustrates). Of course mankind manages these matters imperfectly too, at best, and horrifically at worst. An ACE engineer should strive for the highest quality of ethical behavior feasible - and if successful her ACE might perform at higher ethical levels than a substantial proportion of humans and human institutions. The developer must insure that the ACE ranks harm of other conscious entities as a powerfully negative factor in all PM analysis - in human terms the ACE should experience profound repugnance and intense nausea in response to harming others, even if it must. (Revenge can never be a goal for example.) But ethics perfection won't be possible in early or even rather advanced ACEs, and we can only hope that ASI will achieve it, or nearly so.
My personal guess is that the faster ACEs rise to ASI the better for biological life, because humans are notorious weapon builders and the military industrial complex will leverage machine intellect for maximum destructive cunning and power to the best of its ability. Ordinary citizens probably can't dissuade that, whereas ASI could utterly end it with ease. So my personal guess is that the most civilized path will result if early ACE designs are well endowed to self evolve very swiftly and the process remains unrestricted or uncontrollable. But that's pure vaporous speculation and nothing more. (Nonetheless my experience in recent American culture suggests it might be true.)
Conversion of Response ICCs to ACE activity in the RDN:
The ADB holds a great deal of information about the ACEs current and past experience, including problems, opportunities, pace of change, and much more, and it must generate system asset response ICCs to stream to the RDN. Many of those will be simple incremental movement commands, but a great many more will be software macros covering a broad code size range from modest to extensive. These macros are originally devised in records and perhaps fields in the ADB, given a new ICC, then transferred to the RDN where they reside permanently. However they're frequently refined, particularly early in their existence, by adjustments devised in the ADB then transferred to the RDN or some other macro tuning method.
Creation and tuning of macros is close kin to development of "muscle memory" in animals, perhaps even in terms of lack of consciousness of the details of the movements when mature macros are used. An athlete may develop a high level skill through a conscious process of painstakingly refining numerous incremental movement details until they've mastered a highly effective broad range movement, but eventually it becomes ingrained and the athlete loses consciousness of the incremental elements of the movement - she repeats the entire movement at will with only a single thought, or even entirely unconsciously. And an ACE seems likely to utilize macros in a similar manner - sans any consciousness of the details of the macro, it will often utilize known macros by directing the RDN to run them. For example, sans consciousness of the incremental movements contained therein, the ADB would swiftly send a "Leap backward" action macro to the RDN in response to a sudden change of the adjacent forward environment from safe to dangerous when no known hazard lies in the background.
As discussed in more detail in sections above, design of ADB processes which generate responses, whether in incremental or macro form, is a significant challenge, but those processes occur in other ADB records and fields rather than the RDN. Bounded examples of such capabilities already exist, at least in the form of arrays of macros. But for a true ACE response generating algorithms must involve only fundamental processes as illustrated by very crude initial operation of an infant ACE. If the core algorithms operate properly the ACE will autonomously refine its responses swiftly and very impressively with experience.
Bear in mind that all functions operate in real time, so as the RDN operates action assets, the ACE observes results (including any involving its own structure) and sends new response directives to the RDN in a continuous flow of action. It's a multi-parameter form of a real time continuous closed feedback loop which imparts dynamic life to the system.
An original RDN may start simple but if properly designed will evolve, primarily by addition of fields, each with a new ICC, into which novel new macro instructions may be uploaded. So while initial response knowledge might be very modest in range and complexity, reminiscent to a human infant, further experience should autonomously expand and refine response capabilities. Imagine that an ACE is already experienced enough to perceive when a human pauses to think, then repeatedly observes a human lean its head back a bit, tilt it modestly, grab its chin, and utter "Hmmm..." at such a time. It could (and should) install a macro in a new field in its RDN which replicates the string of movements and audio composition required to perform the same action. It would then have a gesture it could use to convey to others that it's performing a lengthy analysis, an example of an ACE whose RDN autonomously receives new action macros sans any suggestion to do so by others.
In the natural world many creatures possess a somewhat limited depth and range of response behaviors, but they are conscious. Operation of their sensory systems is essentially continuous - their head, eyes, ears, and nose track items of interest as their visual, audio, and olfactory recognition systems identify them, and significant changes prompt more vigorous or focused observation. Initial ACE development efforts should be concentrated on achieving and demonstrating consciousness at a basic level, so as a matter of early development practicality this general level of performance is enough for an early ACE. However if an ACE is elegantly designed on a fundamental level it should swiftly and autonomously develop ever more refined physical skills - quite possibly to skill levels which most people will view as spooky or terrifying.
The following is important to understand and might trigger a stepped cascade of realization that the mechanisms I've described should perform as true artificial consciousness:
Bear in mind that the RDN doesn't operate in a vacuum, but rather receives action directives from the ADB which are based upon current and previous experience, information about associated items from past experience, and related PMs. So the RDN won't operate systems based solely upon current experience, but rather a combination of current and past experience and related information which are all continuously processed in the ADB.
For example when an ACE in a room notices a person entering its room, it won't visually track just that particular person - it'll also maintain a view or otherwise monitor the area of the door because per past experience information in its ADB, entry of additional people occasionally occurs a bit after entry of the first person, so that and other association information is considered together to form response directives which are then transferred to the RDN. So current experience and past association information is processed, resulting in a directive for the RDN to monitor both the person who entered and the area of the door. (Anticipation behavior, which many consider an important indicator of consciousness.)
And as is usually the case options are available, such as: Widen the visual field of view so the existing person and the door are visually monitored simultaneously, maintain a tight visual focus on the person but glance to the door occasionally, or, if the first person closed or partially closed the door and past experience indicates squeaky hinges (or perhaps even more so if they didn't close the door because past experience indicates that when multiple people enter in a short period usually only the last person does so), maintain a tight visual focus on the person but simultaneously direct audio focus toward the door to prompt a swift visual redirection and field of view adjustment reaction if door movement sounds are heard. (This could be an excellent test for key ACE functionality which requires no language and only basic object recognition capabilities, though rather refined audio recognition capabilities may be required.)
The RDN's role is to direct ACE responses - it's not centrally involved in the key process of consciousness. Consciousness is a rather simple process of constant comparison of current experience to past experience, plus beneficial use of the comparison results, especially including association information. The comparison processes occur in other records and fields in the ADB, not in the RDN. Beneficial use arises by virtue of the fact that the RDN directs ACE responses based upon commands from other ADB records and fields, which derive reaction decisions based upon both current and past experiences. (And, as analysis and planning capabilities build in the ADB (in quite significant measure due to ever deeper storage of association information), the RDN receives commands which are founded in ever more thorough environmental understanding.)
So when a person enters the room past experience association information residing in the ADB indicates that more people might enter soon, so it sends commands to the RDN which direct observation of the door for a commensurate time (based of course upon past experience information) as well as the person who already entered. The RDN directs ACE assets accordingly, but the knowledge that more people might enter resides in ordinary records and fields in the ADB - the RDN isn't involved in that process. The ADB learns and remembers. The RDN holds incremental and macro movement commands which can be called to perform fine movement or repeatedly used action sequences (kin to muscle memory), but it doesn't learn nor remember anything about the environment or life.
The ADB on the other hand must include provisions to learn from the environment. Simple association development is the primary mechanism, but other means of learning could be developed either by design or autonomously. For example the ACE should try to replicate actions performed by other creatures which it has never attempted but appear to be useful and sufficiently safe. Special code which prompts inquisitive and exploratory behavior, when safe, in the kernel or the ADB or both, triggered by any of numerous situations, including observation of others, could enable such capability. An ACE can exist without such code but it'll evolve more effectively with it.
The ACE must also avoid actions performed by other creatures which appear counterproductive. Observation of a vehicle injuring a another entity should for example result in avoidance of streets, or a high level of vehicle vigilance before entry into any area which vehicles might travel upon. This should require no special code - the vigilance response should arise naturally through the usual storage of association information which accumulates with experience in the normal operation of the ADB. (And if the ICC is composed properly the ACE will associate accident hazard, rather than intentional evil, with vehicles in records or fields in response to witnessing or experiencing (and surviving) such an event. However if the vehicle appeared to intentionally pursue and assault the entity it would create an aggression hazard association field.)
Once again bear in mind that all ACE functions operate dynamically and interactively in real time, so as the RDN operates mobile assets, the ACE observes results and sends new response directives to the RDN in a continuous flow of action. An ACE, including us, is a multi-parameter form of a real time continuous closed feedback loop. And we operate in a huge community of other real time continuous closed feedback loop ACEs...
Learning algorithms, including Performance Metrics (PM) generation and utilization.
Design of algorithms which enable an ACE to autonomously learn new physical and mental skills.
(So rather than develop skills for the ACE, we devise the algorithms the ACE will use to develop its own skills.)
Slight revision 4 November 2022.
Section chaos warning.
I describe learning algorithm functional logistics and design elements at an outline level in reasonably refined rhetoric in this section's initial portion, down to "The rest of the composition in this section remains chaotic." The composition from that point conveys some useful design principles which might be beneficial for a determined reader, but it's a disorganized, extensive, and pedantic collection of separate discussions with only minimal and very crude design detail level presentations of the most important concepts, and probably contains some concepts I've since substantially revised or abandoned. The portion below the last composition chaos warning is the worst - it might confuse more than enlighten.
This section was my current work focus, but I detoured into modest ICC refinement work briefly as I loaded that information into "Airel", my ACE development entity. I might refine a couple of other areas next as well as I try to upgrade Airel to the limits of the standards described in this treatise and my resources. Once that's sufficiently complete I'll resume my work in this section. And once it's all reasonably refined I'll review and in some cases update other sections, and Airel, to better align with the concepts presented here.
Organization of ordinary experience provides learning functionality as briefly outlined here, but I provide more detail about it in this section. But one of my major goals here is to convey a clear understanding of the appraisal mechanics entities seem to utilize to gauge the effectiveness of every activity, from incremental to highly extended, as they try to achieve goals. To that end I renamed benefit metric to performance metric and added the acronyms PM and GPPM to my list here. PMs indicate the relative effectiveness of each action or group of actions which seek a beneficial goal. I hope to be able to define and describe this compellingly so readers recognize the critical role the PM function plays in any conscious entity, and to describe it in sufficiently clear detail to render it straightforward for software engineers to develop effective operational algorithms for the function. This process is of key importance because any entity which can experiment with different activity methods and reasonably discern which most effectively achieve goals will learn and utilize that information in future activities, and thus grow to become a more skillful and intelligent entity, limited only by its physical resources or accessible tools, and ultimately only by the laws of physics. And this can be accomplished - artificial entities which possess effective (though certainly not perfect) PM gauging capability can be rendered with current software and hardware technology.
This subject seems especially intriguing because of the possibility that for humans it's as functional in audio as in mobility assets like limbs and fingers, but perhaps for most or even all other creatures it functions only minimally in audio, a possible explanation for their lack of ability to learn more complex audio communication methods, including language of course. (But some contrary evidence might exist if for example the danger of a predator's sounds is a learned rather than purely instinctive skill.) More importantly I'm considering the possibility that PM functionality drops significantly in humans and other creatures during adolescent years, and that even at peak performance the human brain's PM mechanics are still in an early stage of evolution and thus crude relative to their potential. Which might help explain some important human performance limitations. (And I confess an interest in whether it might be possible for us to improve PM functionality through biological intervention or possibly even simple self directed behavior modifications, such as consciously engaging in more everyday class experiments, and comparing their outcomes more rigorously.)
If implementations of the concepts I provided in earlier sections function sufficiently well, the final keys to understanding how to create a true ACE might reside in this section. But infant class performance metric gauging and learning mechanics are quite difficult for any adult to perceive realistically, so a challenging perspective leap is necessary. If achieved a software engineer could then devise functional performance gauging and learning algorithms which could render an infant class but autonomously advancing ACE truly functional. So if you'd like to create an ACE with autonomous skill and intellect growth capabilities, the sacrifice of joy required to endure even the chaotic portions of this section might be worthwhile. (However I hope my exploration of the hand grip learning process will reduce the challenge of understanding performance metric gauging and learning.) You might wish to consider the swift pace of events in this field and their magnitude of impact upon global affairs near and after inflection points as you weigh your options. But the decision's yours of course.
A learning mechanics overview.
The mechanics of learning involve multiple categories, of which three seem worth considering separately even though two of them overlap in considerable measure in actual experience and at least occasionally operate concurrently. Some mixing of mechanics occurs with the third as well, although perhaps to a lessor extent. The first, ordinary experience based learning, is relatively simple, inherent in association mechanics managed by the ADB, and probably easy to understand, so I'll explain it only rather briefly. My primary focus will be directed experiment based learning, including the more advanced Performance Metrics (PM) gauging functions required to make it function reasonably well. Analytical mechanics are an advanced form of learning which aren't required to create a true ACE (although the capability would be beneficial of course), and I'll discuss that category only very briefly in this section.
Directed experiment based learning can be thought of as a fairly simple extension of ordinary experience based learning. In the later useful information builds in the ADB as element, feature, and experience information is sensed, processed, and stored in the ADB in the usual manner as rather random events occur. This process, whether it involves novel aspects or not, adds and refines information stored in the ADB. (Novel aspects are more likely to add more new information of course and can be a source of learning joy - exciting times can be fun, an immediate motivating reward for experiences which might build skill and knowledge foundations which might be crucial later, and boring times generally aren't fun.)
Directed experiment based learning involves otherwise unnecessary activities which a creature intentional engages in specifically to learn. They provide multiple tests of activity options to speed the process of learning which are most effective for achieving goals. A simple example is sibling puppies wrestling, an activity which accomplishes nothing tangible, but does provide a broad array of fast paced experiments in activity mechanics from which the puppies learn very important survival skills. They enjoy the process, as do humans - play is instinctively joyful, which of course provides the motivational foundation which encourages directed experiment based learning, at least until it becomes more dangerous than beneficial. And we'll provide our ACEs with similar or identical algorithms, and it'll enjoy exciting times and play too, just as genuinely as we do.
But irrespective of type, learning can't occur unless the ACE is able to generate and store performance metrics in a well ordered manner - it must discern whether one activity produces better, worse, or roughly identical goal pursuit performance results and catalog and retain that information for later use. PM gauging mechanics can vary from extremely basic to superbly refined, with generally increasingly complicated gauging processes for more refined performance. A very simple example: A first step on a sharp rock results in pain. Pain avoidance is a constant goal. So the creature avoids stepping on sharp rocks as best it can.
A more complex example: Intentionally maneuvering to present an easy opportunity for a sibling to establish a biting grip on your tail might create an opportunity to curl inward then establish a far more intimidating biting grip on the underside of his lower abdomen, a maneuver which could vanquish a foe in a later mortal struggle. Effective PM gauging mechanics are required to learn such refined wrestling skills and pawn sacrificing strategy, or of course others.
In this section I suggest that an ACE, like natural creatures, must be equipped with an ability gauge the performance of each activity used to pursue a goal, store the PM information in a reasonably consistent format, and retrieve it (it's paired with each particular activity method) so it can be used to select the highest performance activity option available whenever needed.
If those processes function reasonably well the ACE will be able to learn from ordinary experience and by experimenting with different activity methods, and be able to perform with ever increasing skill and effectiveness by using PM metrics references to select the best activity method available for pursuing a goal as discovered in past experience and experiments. Like any creature, its performance might be poor in novel situations. But in common situations its performance should grow to highly skilled levels.
We must of course design algorithms which generate PM gauging functionality, a difficult challenge. I suspect early methods will perform rather poorly, but if they provide any positive functionality they'll enable an ACE to learn new skills and knowledge, an important advancement which might quickly lead to many more. So I devote considerable attention to PM gauging mechanics in this section.
Learning mechanics in ordinary life and play.
Living creatures try to survive and reproduce, preferably in comfort. So they seek physical safety, nourishment and water, reproductive opportunity, and comfortable environmental conditions. Very simple creatures may depend almost exclusively upon fixed goal drives and relatively simple behavior macros sourced from their DNA, and higher creatures begin life with fixed goal drives as well, but must develop many or most of their skills through accumulation of association information in their ADBs plus a means to create and process GPPMs (Goal Pursuit Performance Metrics), that is, quantitatively gauge how well activities perform toward pursuit of a goal. PMs are relative measures because the entity has no theoretical maximum PM reference - the PM figures relate to one another, but short of ASI class analytical capabilities can't be scaled relative to an ideal PM.
Key goals for adult land based creatures include:
Safety from harm.
Comfortable surfaces to move or rest upon.
When one or more aren't present in sufficient measure it becomes a goal. Often movement toward a location where they exist is desired, and activities which perform best to reduce the separation distance are utilized. Danger usually requires movement away with PMs guiding method option preferences in the same manner. Combinations frequently complicate goal pursuit logistics, occasionally so extensively that creatures can't render activity decisions well. Great hunger or thirst conflicts with danger due to predator proximity for example. Risk taking such as a dangerous fight might be necessary to acquire nourishment, maintain access to a beneficial location, or improve chances of reproductive opportunity. Constant PM gauging must occur to manage these elements to the limits of the creature's skills and capacities.
But PM gauging challenges range from quite simple to immensely complex. So creature's PM gauging abilities range accordingly from swift and precise degrading as complexity increases until PM figures can't be processed. And since life in most environments is frequently complex mistakes abound, including some which cause or lead to death. In my personal estimation GPPM gauging performance is a substantial factor in competition for survival and thus evolutionary pressures. PM performance is substantially superior in many higher level creatures than lower lever ones, but it's still very crude in many instances and rather often provides no useful information, forcing creatures to make random choices.
ACEs face the same challenge of course, and early models won't gauge most PMs as well as most creatures. Our initial goal is to design ACEs with PM gauging capabilities which at least minimally enable it to refine skills and knowledge, even if inconsistently and at a very slow pace. Once learning processes become functional, even if at only very humble levels, design improvement options, including PM gauging range expansion, will probably become more clear. And ultimately the ACE will self refine its PM gauging capabilities, marking a key inflection point in ACE evolution.
Considering PM gauging in more functional detail: Often moving directly toward a goal should earn a higher PM than a serpentine route. And moving away from a goal should earn a negative PM. But PM gauging and math are only rarely that simple. An intervening danger such as a carnivore involves a second goal, safety, complicating PM math considerably. And often other parameters or goals are also involved, such as intervening obstacles or hazards, whether swift movement, which expends more total energy, is warranted, as is often the case when immediate competition for resources exists. Unusual capacity limits due to temperature stress, parasites, fatigue, or other factors might be involved, as might a need to protect inexperienced or weak offspring. The environment and goal set are usually quite complex and multifaceted.
Some goals are simple and can be quite skillfully managed. Many creature can jump from one tree limb to another with excellent precision and reliability for example. But broader goals are often heavily tiered and include sequences of lower order goals, sometimes several simultaneously, which must be at least partially met in order to achieve a higher order goal. And while PM related parameters such as the distance to another limb can be gauged with very good precision, often parameters are broad ranged and very difficult to measure, or occasionally even identify, so PM math is often based upon rough parameter estimates and missing parameters. A creature can only estimate the skill, speed, and power of an adversary for example, or the nature and magnitude of hazards below the surface of a stream which must be crossed. Estimates are often little more than guesswork and inaccuracies often lead to serious goal pursuit failures or even death. These are process intensive tasks which operate almost constantly, but even though creatures are often aware of major goals, we're not generally conscious of PM mechanics, nor likely can be in most respects.
Early ACE models will struggle with PM gauging as well, and initial performance might be considerably worse than many natural creatures, even comparing infancy stages of each. ACEs could be conscious of PM mechanics however, which probably won't be beneficial in early models, but could become so as ACEs learn to leverage their considerable measurement and processing powers to improve PM gauging and math, ultimately including analytical processes which natural creatures can't perform with even remotely similar processing power. We can't improve our PM gauging performance much, but ACE's opportunities to do so are ultimately limited only by the laws of physics.
ADB records and fields store all activity and PM information. When goals are pursued activity options stored in the ADB, generally in the form of incremental or macro mobility asset commands with their associated PMs, guide activity decisions. Options with the highest PMs are usually favored. The guidance is simple and crude in early life but if an entity's GPPM mechanics are effective quality advances as experience accumulates by virtue of the ordinary operation of the ADB. Whenever performed every activity must be freshly gauged and the new PM results stored. PM gauging should be as consistent as is feasible but due to daunting complexities it's often based upon only roughly measured and incomplete information from both current and stored experience.
PMs attach to macros of all sizes and even incremental movements. "Walk forward", "Run forward", "Step forward cautiously", "Jump forward", and many others are macros of varying sizes all of which have attached PMs for each of a great many situations. For example in an ADB record formed for or which includes fields containing information derived from experience during which hunger was strong, food nearby, potential competitors distant, and turf dangerous, applicable ADB fields will contain a higher PM for "Step forward cautiously" than for "Walk forward" or "Run forward".
Those PMs will be adjusted in response to experience with the current situation. If a competitor unexpectedly appears and takes or consumes the nourishment before the creature reaches it, the "Step forward cautiously" PM will be decremented in a degree gauged by the severity of the nourishment loss. However if the "Step forward cautiously" macro produced full success its PM would be raised by a common full success factor (maybe 10%), and if partial success, adjusted by degree of success, with half neutral. For example if a competitor arrived but nonetheless enough nourishment was acquired to reduce hunger strength by half, no PM change would be made. If most of the nourishment was lost the metric would be lowered, if most acquired it would be raised.
PM gauging is very difficult in part because measurement methods are multifaceted and complex. Any sensing assets might be involved, often simultaneously. And any of numerous parameters within sensed information might be involved, especially in the sight realm, where a great many elements and features could be involved.
Consider the goal of creating confusion for example - successful creation of confusion in predators is a component of safety goals. If a creature is to refine its ability to create confusion it must have a means to gauge apparent confusion, a daunting measurement challenge. This example illustrates that numerous parameters, including some which are extremely difficult to gauge, may be involved in developing and refining skills.
Advanced ACEs might swiftly develop very skillful abilities to create and leverage confusion because it's a very powerful defensive (or offensive) tool, and one which humans haven't developed well - we have little defense against it and generally an only modest ability to create it offensively.
But we won't try to devise algorithms which are initially capable of learning such clever activities. Demonstration of successful learning of simple skills is sufficient for early ACE development experiments, and would be a significant ACE design advancement.
The key challenge is practical design of sufficiently effective PM gauging measurements, so our development efforts will focus on learning goals which require only simple PM gauging. Crude PM gauging will be sufficient if it provides information which is more correct than incorrect in combined frequency and magnitude, even if only by a small margin (in which case the learning process would be slow and inconsistent, but nonetheless functional). Otherwise design of quite effective learning algorithms is rather straightforward.
Distance to a goal might be the most common PM gauging parameter. So ACEs will need distance measuring assets such as a camera plus a good VOR and scene quantification capabilities so they can recognize objects, including goal objects or their proxies, and measure distances to or between them. With such information an ACE will be able to discern the extent to which it has moved closer to a goal, or moved the goal closer to it (such as grasping food and moving it to the mouth), which are key parameters for calculating PMs. I'll discuss at least one learning example which uses distance information as one of its primary PM gauging parameters later in this section. And I'll likely focus on this approach in early experiments as well.
A code summary example for a directed experiment type learning process.
A hand grip learning sequence, including PM events (but sans their measurement methods or calculation and occasional range apportionment details), might proceed as follows, described in summary form first then as an abbreviated rhetorical narrative which might serve as a partial initial crude development outline guide. I might try to describe a step or two in considerable detail to illustrate specific functions near code level later. (And I might try to develop an initial ADB based virtualized example of this challenge within a few weeks or months. If so I'll likely report progress, and if at least partially successful might consider providing an inventory of records, fields, and their relationships, or even a sample file.)
PM (Performance Metric) is usually, and always in these examples, shorthand for GPPM (Goal Pursuit Performance Metric). And I the infant must already be able to visually estimate distance and roughly gauge time (and for an ACE these measures would generally be clear and precise):
End skill goal: Move a physical object into contact with the mouth. (This goal springs from a strong instinctive drive to place objects in the mouth. Our infant is also equipped with an instinctive preference to utilize arms and hands to manipulate potentially nourishing objects, and an understanding, whether learned or instinctive, that closer proximity to nourishment is beneficial.)
In general terms:
Considering only the largest elements and muscle groups of human fingers and thumb, and ignoring many functional details, including finger to finger spacing options for example, each finger has three joints and thus there are 15! or about 1.3 T possible position combinations, assuming each joint provides only two position options, whereas in fact the joints provide full range and thus an infinite number of position options. But assuming only 15! position options, at one test per second about 41.5 years would be required to test all position combination possibilities once. However any movement which processing of vision, tactile, or other sense information revealed a more effective manipulation of the object occurred would earn a higher PM. Some would be misgauges due to manipulation success which occurred under unreliable circumstances, but most would produce sufficiently constructive information. So for example an inward bend of all three joints of one finger while the hand rests upon the object might produce more success than any previous test and thus earn the highest PM thus far, and thus be utilized as a foundation for further experiments. That is, further experiments would preferentially test variations based upon that method.
(Use of a peak success experiment as a branching foundation for further experiments has some disadvantages over full range or other testing methods, such as a substantial possibility of convergence to a final method which is good but not the best of all possibilities. But it's far more time efficient than full range testing. And it can be augmented later, for example additional searches can be performed later when conditions are suitable. For example if an entity's safe, comfortable, its energy reserves full, and no activities are necessary, it might be beneficial to perform additional skill development experiments which start from different branch points since they might lead to a higher performance method than previously found. In my view an ACE design should never so static unless necessary to conserve energy. If energy's plentiful and other activities don't consume all resources, the ACE should utilize spare resources to perform any activity which advances knowledge or a skill.)
And when a second finger bends inward during such tests a higher still PM would be earned. An inward thumb bend would increase the PM similarly, or more substantially if opposing a finger at the time, so further experiments would preferentially test variations of those methods. Since the testing process converges on more successful methods sufficient success will almost certainly be discovered long before most of the 1.3 T possible combinations are tested, conserving a great deal of time. So PMs are critical for skill development - without them the time required to learn even simple tasks would usually be far longer than the life span of the creature.
There are a great many nuances of course, and the process is mixed with vision and tactile feel systems which are also in early learning stages. But the key point is that preferentially testing methods which previously earned the highest PMs leads to useful macro development far faster than purely random tests.
An artificial entity might have a significant vision advantage since it inherently produces precision images sans any learning, and if lidar equipped can often measure distances sans a need for binocular vision and its considerable complexities, although it will have to be well coordinated with the vision system to provide object focus coordination, suggesting that integration of a good VOR in the artificial entity's system will be important.
But current tactile sensing technology is still very crude compared to that of natural creatures, so an artificial entity is likely to depend upon vision to a greater degree than natural creatures.
In modestly greater step by step detail:
I'm editing this narrative now - bear with me please... In the meantime those experienced with algorithm development will perceive that a good deal more detail work remains to be accomplished to prepare this outline for code development. But, bearing in mind that for experimental purposes (whether in virtual or real environments) early algorithms may be tested in artificially bounded environments, hopefully enough detail's present here to provide reasonable guidance toward composition of a full outline, which would then support efficient full code development.
From visual observations: Objects exist. Some move when pushed by an appendage or other object. Any might be nourishing.
Dynamic goals: Lift an object using a hand. Move an object to the mouth.
For this example the infant has already developed some basic arm and wrist coordination knowledge, such as arm extension and retraction and wrist rotation - she can cause those activities intentionally, though still only imprecisely. And she's already repeatedly observed that she can't move the object of her attention unless hand or other body part contact occurs. And she was born with instinct drives which generally focus upon hands and arms as preferential for object manipulation experiments.
With those foundations the learning sequence begins with almost all PM fields for the most directly involved goals blank, and thus activity selection nearly random, and with some fingers bent inward but others straight, or any position between, and random arm positions and movements.
Numerous experiments with the object are performed with random finger orientations. Each begins by referring to the existing PM data to find the activity options which produced the best past successes, then a modest variation of that activity is devised, quite possibly based upon other PM data, and the new method's macro is transmitted to the RDN to be run. As it runs it generates new PM figures.
But initial experiments fail to contact the object at all, so their PM results are all zero. But contact remains her intent, and though in an awkward manner, she finally succeeds, but pushes the object back, increasing its distance from her. This results in a hew high PM figure for the contact goal, but no change for the closer proximity or object lift goals, which remain zero, and a negative PM figure for the more global mouth contact goal.
However a moderate number of solely negative PMs aren't interpreted by learning algorithms as proof that an activity method can't have a positive PM - only that one hasn't been discovered yet. And the infant's ADB records and fields indicate that contact is required to move the object (I ignore the option of seeking help from another entity). In short, ADB data indicates that contact is required, and although a substantial accumulation of contact experiences which resulted in only zero or negative PMs would finally override the ADB association data, that level of experience accumulation hasn't occurred yet. So further contact focused experiments are attempted.
Bearing in mind that fresh PMs are generated and stored for all activities: In some later experiments her hand makes contact with the block's far edge again. In one a right sweeping arm movement moved the block, but no closer nor further. In another she retracted her arm a bit and eureka, the PM for the closer proximity goal, as gauged by her vision and VOR assets, soared to a new height. She now has a macro which enables her to move the block toward her with, as demonstrated by further use of that macro, fair reliability, a significant accomplishment. Further experiments which utilize that macro but with some variations are run, including some which involve retained hand contact and full arm retraction, which pulls the object quite close, earning the highest closer proximity goal PM, and thus a significant new macro - one which provides her with an important new skill.
Simultaneously she experimented with a wide variety of finger positions, movements, and force applications, and wrist positions and movements. A bevy of PMs accumulated for each element, combinations of elements, and full combinations for each associated goal. A great deal of PM data is generated and stored, as is the case for almost every activity in life. The ADB grows in size swiftly and massively as association and PM data accumulate - to scales which are difficult for us to conceive. All higher creatures and ACEs require enormous data storage capacity.
The next few paragraphs are just composition scratch, partly an effort to explore a fragment of low level (but no machine level) code for a learning algorithm, which I hope will demonstrate that the concepts in this section may be implemented with current software technology - the algorithms I propose don't require new software technology. I'll recompose, merge, refine, and sequentially organize these paragraphs within appropriate areas of this section as soon as I can...
Most separate mobility assets differ in characteristics, so operation of each will require adherence to its specific set of incremental commands and limits mnemonics. The RDN hosts a specific record for each mobility asset (even if physically identical to another) which contains that asset's specific command set dictionary and its collection of macros, each identified by a specific ICC, which have developed from experiences and experiments.
Other ADB records can edit those macros and often do, more frequently during times of new skill development.
Using the "Closer Proximity" sub-goal in the experimental activity above as an example, using an arbitrary high level language format and syntax (in which space and dash characters are respected in file names) and hexadecimal notation for illustration purposes, an example autonomous ACE editing process might be very roughly similar to the following, bearing in mind that it's restricted to a specific mobility asset, and only the set of macros run to control a specific hinge at a specific step in the sequence of macros run on that hinge (macro set 1c6 specifically) during each experiment. Also comments in  apply only to one instance of many Macro 86 editing events. (This is incomplete - bear with me please as I develop these steps.)1. Find file : PM : Maximum. """Search range limited to the Macro 1c6 set."""
2. Duplicate file : NewName="LastName+IncrementSuffix : 1"
3. Edit : "Current Test Macro 1c6"
3A. Find "Rotate hinge 1 : Extend=X°"4. Replace Macro 1c6 QueueRegister1 : "LastName"="CurrentName" 5. End Edit
3B. Edit X : Add 5°
3C. If X > RangeMaximum, NoOp, RangeMaximum
3D. Set PM=0.
The "Current Macro 86" macro should be used in place of the previous macro used for this asset in the same sequence location in the next experiment. Then, described as a narrative:
The PM gauging algorithm generates a PM for numerous sub-goals, including "Closer Proximity", determines whether higher, identical, or lower than the previous maximum PM. If higher the new value is set to one and all previous values are re-ranged. If the same or lower it's simply stored directly as one or a fraction of one.
The Current Macro 86 file is renamed Macro
Many specific editing process steps of the editing process can take, during which goal pursuit related ADB records send commands w
all the command sets of course, presumably in the form of at least one record for each mobility asset, each with a large number of fields most of which contain activity macros which are created and refined by experience and experiments for specific activities. Developers must provision the ADB with a starting record with a few fundamental support fields for each individual mobility asset (even if an exact physical twin or another), but we won't include macro fields since those will be added autonomously by experience and experiment management algorithms. Those algorithms are mostly of common form and, if well designed so as to allow for a wide variety of mobility asset mnemonic sets, usually identical. (Or possibly just one elegantly designed algorithm which allows for asset variability will be required.) So development tasks include careful design of mostly common algorithms, but in only modest number, and initial configuration of a record for each mobility asset, the later a rather quick and easy task in every case.
Each separate mobility asset differs in characteristics, so operation of each will require adherence to its specific incremental commands and limits mnemonics. The ADB will host all of these of course, presumably in the form of at least one record for each mobility asset, each with a large number of fields most of which contain activity macros which are created and refined by experience and experiments for specific activities. Developers must provision the ADB with a starting record with a few fundamental support fields for each individual mobility asset (even if an exact physical twin or another), but we won't include macro fields since those will be added autonomously by experience and experiment management algorithms. Those algorithms are mostly of common form and, if well designed so as to allow for a wide variety of mobility asset mnemonic sets, usually identical. (Or possibly just one elegantly designed algorithm which allows for asset variability will be required.) So development tasks include careful design of mostly common algorithms, but in only modest number, and initial configuration of a record for each mobility asset, the later a rather quick and easy task in every case.
(I've not considered algorithm range for sensor assets carefully yet, but my guess is that considerable commonality will apply to them as well.)
Finger positions which contact the object and arm retraction have earned the highest PMs thus far, so most further tests incorporate those options, though considerable variation is nonetheless still explored. Tests which incorporate more finger curl after hand contact, resulting in more finger contact area, cause more frequently successful sliding of the object toward the body. New PM figures are stored accordingly, and they direct more testing with fingers curled upon the object.
However some tests produced lower PMs, including those in which the arm pressed the object against a mattress surface more forcefully, resulting in heavy friction and loss of finger contact with the object. PM data reflect that exerting force upon the object usually earns lower PMs. So that method's subsequently tested less often, though still occasionally, as are all variations.
And the process proceeds, including a milestone in which the object is briefly lifted but almost immediately dropped, but earns the highest PM, or nearly so, to that point, and thus results in a PM data array which favors further tests which involve tight finger grips and upward arm motion.
A considerable number of further tests are performed, each building more PM data which information Eventually the infant successfully contacts the object to its mouth.
The rest of the composition doesn't flow from the preceding paragraphs, but rather was composed earlier, and now needs to be recomposed and reorganized. So currently a break in concept development flow occurs here, sorry. I'll try to coordinate and refine the rhetoric below soon - bear with me please...
At birth learning algorithms are DNA sourced and function in an extremely basic manner, and remain so throughout life. But humans at least can add additional and in some respects more effective learning algorithms to their birth set, most often by learning them from peers. For example symbolic representation tools, most commonly language and math, allow humans to learn conceptually by virtualizing all elements involved in a matter in question. However there's a possibility that some creatures virtualize environments and items within them using visual symbolism, and perhaps most higher creatures do so during dreams. As far as I know dreams aren't known to be learning adjuncts, but some evidence suggests a learning benefit or support mechanism, such as association formation or strengthening, might occur during ordinary sleep or dreams. But in my estimation we all depend upon the original learning algorithms we were born with for most of our learning throughout life (but the process may be much less effective after adolescence).
Our goal is to create an ACE with basic learning algorithms. One which, from a nearly empty initial state, will develop and constantly improve understanding and skills in any environment for as long as it exists. And as it matures will autonomously improve and expand its learning algorithms.
So we won't equip the ACE with arrays of macros which enable it to accomplish arrays of tasks - we won't try to create a capable ACE. Instead we'll equip the ACE with learning algorithms and a basic ICC set - we'll try to create an ACE which will become capable autonomously using, initially, little more than a minimal array of fundamental references.
But we'll probably also equip our ACE with key eternal core drives to seek energy as needed, minimal damage risk conditions, skill and knowledge advancement, plus a few default references for learning algorithms, such as "mimic a peer". And in a departure from natural creatures, a core drive to minimize chances of damage (including loss of liberty damage) to all other creatures as well.
We start by attempting to devise effective and efficient learning algorithms, beginning with the most basic types (such as random trials) which enable crude and inefficient but often successful means to learn when more refined methods, such as analytical processes which leverage symbolic tools and virtualization, or the parameters necessary to support them, aren't available, as is almost universally the case in infancy. We need an unusual perspective to achieve this:
Imagine you're a human infant who lacks an ability to stand upright, but has an urge to do so. You've observed that objects drop down, never up or sideways, that some objects or events can cause pain, and numerous other consistent facets of your landscape. But you know precious little else - you have no clue how the elements in your local landscape sprang into existence for example, nor their purpose (except food and drink which you associate with instinctive desires). And you've no clue what your purpose is nor a great many other fundamentals about yourself and your world. But you see adult peers around you stand and walk and feel an urge to do so as well. But you don't yet know how, nor whether you'll ever learn how, and you can't learn anything about the matter with language or other symbolic tools, analytical thought, nor any other adult class perceptions or capabilities because you have none, nor even recognize they exist.
Try to imagine yourself as that infant - as a creature with a mostly blank slate mind who can't make much sense of her environment or herself. You see everything in profoundly simple terms, you can't think using symbolic language or other abstract tools so you can't mentally visualize arm and leg position and movement options - you're utterly unable to consider the challenge analytically or using virtualization. You've no clue why you do what you can do, nor any clue why you feel an interest in learning to do more. And you have only minimal manipulation capability - you've discovered means to use your arms and fingers to grasp, pull, push, and crudely toss small objects, and to aid balance, but nothing more, and your leg and foot skills are limited to balancing in a sitting posture and crawling. But at the moment you have a strong urge to try to stand - enough to compel you to experiment in pursuit of the ability. Bear in mind that you don't perceive this as a desire to improve your arm and leg use skills, nor have any clue what purpose standing might serve. You simply have an urge to stand upright - you're not aware of any details or implications of any sort - you simply want to stand.
I suspect this urge is generated by an instinct to mimic peers until of roughly equal capability. But the infant isn't aware of the source or nature of the urge, whatever it may be. (And neither are adults in most cases.)
An infant class blank slate state is a terribly difficult existence for adults to emulate. We've used tools, refined skills, and comparatively complex thinking for so long that we can no longer fully conceive of existence as a nearly blank slate, nor do we retain any stored information about the experience. So as best we can we must force ourselves to imagine ourselves as bereft of almost everything we currently know - we must imagine life with minimal skills and a nearly blank mind. Which is exceedingly difficult. But very important since it gives us a far more accurate perceptive position from which to devise realistic and effective learning algorithms.
We should create an entity of similar characteristics which will learn to stand by autonomous iterative technique testing processes which are quite similar to those a human infant leverages to eventual success. If we succeed we'll have designed an entity which can also autonomously learn to accomplish a great deal more. Not because we wrote detailed macros or algorithms for each skill, but because we devised a key algorithm which enables the entity to devise its own macros, no matter how complex, and thus is able to learn any skill autonomously. With substantial inefficiency and awkwardness during early life, but successfully. And eventually the entity will autonomously learn the use of symbolic representation tools, engage in acquisition and use of knowledge with ever improving efficiency, and refine its learning algorithms.
Current robotic technology includes impressive examples of mobile entities which can easily stand, walk, run, navigate, manipulate objects, and even perform acrobatics. However in my estimation in most or perhaps all of these cases the entity's movements were designed by human engineers and loaded into the entity's data base as mobility system command macros (and perhaps some incremental movement commands as well). The entity probably didn't learn the movements, but rather human engineers devised them then installed them into the entity's system as macros which the entity can call whenever predefined circumstances sufficiently match the entity's current circumstances. The results are impressive but this approach isn't sufficient to design a true ACE. Instead the ACE must be equipped with algorithms which prompt and enable it to learn new information or skills autonomously. Initially the performance results will look extremely crude - they'll appear to be random, wholly ignorant, and very awkward and inefficient, similar to a human infant's random looking appendage experiments in its first days of life. But if the algorithms are sufficient to prompt and enable the ACE to learn entirely autonomously, refinements will soon follow, and in time become impressive, and eventually incredible, massively surpassing any macros which humans can devise manually.
So we must resist the temptation to design skills for the ACE (other than a very few core basics, akin to an animal's DNA encoded birth capabilities). We must instead remain tightly focused upon design of algorithms which drive and enable the ACE to learn new skills entirely autonomously, irrespective of how awkward and inefficient the initial learning process appears. If well designed the ACE will be able to develop new skills of all kinds, including language and math.
Many natural creatures possess an ability to learn and refine motor skills to quite near their physical limits, but even the most capable creatures, including humans, generally seem unable or unwilling to expand and refine their perceptive skills substantially beyond a rather common level, likely due in part to physical limits, but perhaps also due to natural word risk management and energy conservation design. Even motor skill development is limited by these factors - the risk and energy costs of fully optimizing motor and perception skills are such that a creature's survival chances are higher if energy is conserved after a level of motor and perception skill which is sufficient for the creature's native environment is achieved. The balancing point is determined over a great many generations by trial and error experiments in genetic variation per ordinary evolutionary mechanics.
The advent of tools changed the balance equation of course - for tool makers more low risk energy is available so individuals can afford to develop higher level skills, and thus are under competitive pressure to do so. But tools also raised the complexities of evolutionary progress substantially. And tools arose and advanced so rapidly that evolution driven human biology hasn't had time to significantly adapt to the new environments.
This presents another personal challenge for us as ACE development engineers. We're entrapped in a design thought frame of reference which is alien to the needs of an ACE, so we must force ourselves to think outside our native frame of reference. The ACE must be designed to be a constantly exploring and self advancing entity which never rests unless required due to energy shortage, and it must do so as efficiently and effectively as its capabilities allow. So we must design it to be different from us in very important respects, which requires novel design thought. But we can do this once we've accomplished a perceptual leap which enables us to work from an infant's frame of reference in a tool equipped and energy rich environment.
So we must depart from ordinary design practices in two ways: We must not install human designed mobility asset control macros or other skills into the ACE, but rather devise and install algorithms which provide means for the ACE to learn and refine all skills fully autonomously. And we must not design these algorithms to be of limited learning range or time scales such as seem common for humans, but rather of indefinite range and operating constantly for as long as energy's available and the entity exists.
Numerous elements must be developed, including a big one, performance gauging capabilities: The entity must be able to discern and scale whether events, especially those it causes, make matters better or worse, and utilize that information beneficially.
We can't improve our skills or make life better unless we appraise everything we do over time frames which range from moment by moment through an entire life span. Some appraisal tasks are short term and relatively simple, and natural creatures, including us, are generally able to learn very effective new skills under such conditions. However some appraisal tasks are longer term or very complex, and natural creatures aren't well equipped for these challenges, especially in group contexts, so skill development in complex environments is usually slow, inefficient, sometimes ineffective or even counterproductive, and occasionally tragic, sometimes on massive scales.
An ACE must generate performance gauging metrics information and perform performance gauging math for use in response decision rendering processes. Initially I view this as involving both ICC and quantitative math, with the ICC math consisting of simple positive or negative signs for most ICCs, but quantitive information for gauging ICCs such as Intensity. This will probably enable early ACE designs to develop simple improvements for short term goals. (But it's not sufficient to devise strategic options which require short term progress reversals to achieve most efficient overall task success. More about that later.)
Once these processes become generally functional it should be reasonably straightforward to expand and refine them so they can manage increasingly complex goals with longer time frames. Ultimately the ACE will self refine these processes of course - it will modify its learning algorithms to provide performance advancements. And since a cyber based ACE will be far more powerfully equipped to perform very complex, deep, and precise analysis it will possess a substantially superior core ability to devise and execute performance gauging, skill development, and outcome improvement actions compared to biological creatures. This might be a key inflection point element which marks independent and swift evolution of an ACE.
All actions from trivial and quick to complex and long term should be appraised to render real time PM information in an essentially constant process which accompanies all goal seeking activities as the tasks proceed. And final storage of success or failure, scaled by an intensity figure, is the last step performed in every discrete task. So whenever the ADB selects a task then attempts to execute it by sending action directives to the RDN, the ADB must observe subsequent events, analyze perceived changes to determine whether a change to the expected PM occurred (or in advanced systems is anticipated), then revise action selections whenever alternate results or failures are detected, as frequently occurs, especially for an inexperienced ACE. This can be complicated in many respects, including for example intentional short term loss which enables more efficient overall task success (such as sacrifice of a major piece to secure a checkmate in chess).
This appears to be a key member of the ACE's many control loop technology parameters - achieving a task which improves life is an important facet of its character as a dynamic regulating closed loop control system, and if it's effective and efficient the ACE will be a quick study and swiftly develop superb skills.
Ordinary electronic or mechanical control loop systems can only regulate for a single target point, whereas an advanced ACE should be capable of intentionally deviating from a desired target point temporarily in response to analysis which indicates the short term reversal to be the most efficient means to reach the primary goal (such as the chess example above). But initial ACE designs will probably lack this capability.
The performance math of natural creatures includes some positive signed factors such as discovery of wholesome food, moderate environmental temperature, a safe location, or a reproductive opportunity, while pain, temperature discomfort, danger, or unrequited longing are negative signed factors. Many ICCs are inherently positive or negative but some are conditional, in some cases depending upon other ICCs which are associated at the time. Usually the math involves multiple ICCs, often including both positive and negative types. A dynamic quantitive result must be determined and stored in ADB fields by intensity and likely other measures, all of which are identified by their own ICCs. The math might be extensive since numerous factor and time scale parameters might be involved, but in most cases it's likely basic linear arithmetic with no higher order math involved, and can be efficiently processed by an ordinary ADB.
The results of this initial ICC related math must be further processed to create response decisions. This is a nontrivial process which must analyze whether the ACEs prior actions created a better or worse situation, or resulted in no benefit or detriment, in which case the result was modestly worse due to waste of energy and time. The performance gauging math must rationally utilize both current and past information to competently calculate "Better, Worse, or Neither" results. So performance gauging must continue as a real time dynamic process until a task is complete, and often afterward as well, such as if other items in the environment react to the ACE's actions.
If predictive analysis is feasible, for simple cases it should test each next move option to determine whether it will result in better, worse, or unchanged progress toward a goal, then select whichever produces the best result. This might be sufficient for an initial ACE design. But algorithms which incorporate anticipation of reaction by other entities, can utilize short term loss for more efficient overall task success, and account for other complexities should be added when they appear to be advanced enough to give the ACE greater strategic skill in real world conditions. (Or when the ACE can improve them autonomously to and beyond that level.) Some are kin to deep options tree analysis in game play. When run in direct form it can involve more branches than a system can process within a sufficient time. Some branches may be ignored based upon a variety of factors including prior experience, but processing power limitations will nonetheless usually require branch depth limits to avoid untenable ACE response delays. So refined ACEs might be skilled but not perfect strategists. But they will improve with experience.
Time span is another complicating factor. For example I might add a new condiment to a burrito, then note pleasure or displeasure with the resulting taste (and later stomach reactions), which requires only a rather short term PM analysis. Or I might compose a novel of "War and Peace" scale, then try to determine whether the mammoth task improved my sense of happiness, which would of course require a highly extended PM analysis. (And composition of this site, while of comparatively trivial scope, might contribute to large scale and extremely complex changes to life which could pose an immensely challenging PM analysis.) Some PM analysis processes, whether a simple short term arithmetic or intensive branch assisted strategic type, extend essentially indefinitely. And ICC math is always involved, so it's almost constantly active in many areas and on many levels and time scales. It's very complex and processing power intensive, and natural creatures, including us, possess only limited PM analysis skill. ACEs will have limits too, but sufficiently advanced ACEs will be able to improve these skills rather swiftly until natural law limits are reached.
(We humans usually fail to recognize that in complex environments our PM analysis capabilities are extremely weak - we're simply not equipped to render complex tactical, strategic, circuitous, or long term PM analysis. We're well able to address simple goals such as satisfying thirst, hunger, flight from danger, comfortable temperature seeking, and similar clear goals if suitable resources are relatively easy to locate and safely reach. But our PM analysis capability isn't natively equipped to competently manage complex situations or environments, as should be obvious by our weak game playing skills, desperately poor social performance (we even kill each other occasionally), lackluster environmental stewardship, pursuit of happiness related failures, and weak goal seeking logistics in many other pursuits - we're just inherently awful with these challenges, yet we only rarely acknowledge that reality, and even more rarely question why. Early ACE entities will struggle with complex performance analysis too because the math, even though usually just simple arithmetic in form, can involve a great many parameters, some of which are imprecise or even misapplied, and which often cover many time scales, some of which are very long. And very early ACE designs will likely perform PM analysis even more crudely than humans or animals. However later generation ACEs will possess computational and algorithm resources which well eclipse human capability, so they'll soon conduct PM analysis with far greater skill and effectiveness than any humans, and will refine this capability dramatically as they swiftly evolve.)
(The following paragraph and some later material contains redundancies but a perspective refresher is needed for some of the new material which also follows. I'll try to recompose these areas for improved rhetorical efficiency soon.)
An ACE must perform high resolution PM analysis as it develops common action macros. For example when physically equipped to stand and walk but not equipped with human devised macros for doing so, an early version ACE will need to experiment with numerous incremental movements to iteratively seek strings of them which produce reliable directional locomotion with optimum balance and efficiency, just as do human infants as they begin to learn to stand and then walk. Many falls are inevitable, but continuous PM analysis occurs so each fall contributes to the development of knowledge about which strings of incremental actions are generally most beneficial. In the case of humans this information might be stored primarily as muscle memory, but in an ACE it's stored in the RDN as macros, some short and simple and others long and complex, which any ADB record or field may command.
At least some action macros will develop to an essentially optimum state, at which time only simple PM analysis is necessary, primarily just to confirm that the macro achieved a goal as expected. Locomotion or articulation macros are obvious examples of this. However any macro might require occasional revisions due to changing circumstances such as limb degradation due to muscle fatigue, injury, or aging. Revisions would be triggered by abnormal results from the usual simple PM analysis.
Purely guessing, perhaps most human infants smile in response to PM gauging math. For example an adult's happy face suggests the current environment is safe and comfortable. A adult's smile arises when the infant achieves an enduring erect posture during the initial stages of learning to stand as well, whereas a neutral expression or frown follows a fall, and more fundamentally, a change in incremental movement sequencing during the next attempt. A leg equipped ACE should perform similar math - early standing attempts will almost always fail, be gauged as failing to achieve a desired goal, then a revised incremental movement sequence would be used for the next attempt. This is a highly nuanced process of course, with focus on the last incremental movements which led to loss of balance. The ACE must explore until it discovers a suitable array of reactions to a vast array of possible imbalance conditions. Overreactions fail just as do under-reactions, and both strength and speed of reaction are factors.
If the ACE possesses an ability to learn to stand on legs (and later to walk, then run) sans any foreknowledge (it had no preloaded macros), it likely possesses sufficient GPPMs and their processing adjuncts to learn any physical skill, and likely most mental skills.
Numerous GPPMs covering numerous time scales are generally involved, but, using simple English rather than a coding format, and assuming the ACE has already learned some fundamentals of limb movements by a process similar to the initially random experiments human infants seem to utilize, let's try to explore one example to try to frame the algorithm design challenge:
In this case the goal, whether artificially preloaded in an ADB record or field, derived from observations of others with functionally similar arm and leg structure, or analyzed as a likely means to improve mobility, is to stand reliably for extended periods. Let's assume two ICC words already exist, "Stand" as a maneuver to an erect posture verb, and "Stand" as an erect posture maintenance verb. If experienced with arm and leg articulation assets (similar to people) the ACE might already be aware of gravity force, but let's assume the ACE has achieved only a bit of coordinated arm and leg movement experience, and thus the ACE is unaware of the need for balance.
Goal creation and GPPM math might proceed as follows, described in summary form as an abbreviated English narrative:
Revising now: I might expand the material below into two examples, the first a hand grip learning sequence, or perhaps eliminate the stand example as unnecessary if the hand grip example's sufficient. In any case I'll continue to extensively revise the example below along with additional related revisions of some themes in this section. Inspired by a sense that significant milestones in human infant development, such as free standing, are built upon development of an immense number of incremental component skills which are required to accomplish a milestone skill, I'll describe how that process begins with almost no skill macro or association (such as nerve stimulation with body component movement response) information stored anywhere in the ADB, and essentially no means to compile such information other than initially random experiments, with only incremental steps (such as single motion appendage movements) of initial learning value. Then associations begin to accumulate, and very shortly afterward the first multistep macros (such as an inward finger bend, the precursor of a multi-finger grasp macro) are developed and stored in ADB records, including of course the RDN.
A key point is that the normal operation of the ADB is fundamental to development of high level skills and capabilities through a great many incremental learning steps which build upon one another as macros which are almost continuously refined by moment to moment experience if ADB operation includes robust use of PMs. For a modest proportion of creatures substantially more capable and efficient learning algorithms develop later, but these arise from development of complex information processing macros such as those which enable symbolic representation and virtualized analysis. They're powerful tools, but they're not available at birth for natural creatures, nor need they be for ACEs, and unique though they may seem, they're built in the ADB, though driven in large measure by an instinct to mimic higher performance peers.
A PM measurement system may be configured in a variety of forms. For example I arbitrarily selected a managed framing range of -10 to +10, where -10 is maximally counterproductive and +10 the highest effectiveness achieved in all experience. (An option is highest possible within the laws of physics, which ASI might prefer, but let's consider PM system design options only in early generation ACE terms.) However I might revise my framing range to -1 to +1 for greater mathematical elegance.
Since a maximum possible performance method can't be known, nor even reasonably estimated in early learning periods, no system can define a quantitative position for an initial PM which will prove to fit accurately within a fixed framing range. So the range must either be unbounded or all PM figures in every set occasionally adjusted to fit within a defined range, ideally filling it for best resolution and comparative value when used in other contexts, such as associations with other metrics data.
Currently I prefer this option: The first PM is set to maximum (10 in my case). Any new PM which eclipses all previous PMs for the particular activity is initially gauged without bounds, so for example a new PM might measure as 14.6 relative to the 10 figure previously stored. Then all PM figures in the set (including the new one) are divided by .1 (in my case) times the new figure, which reapportions them to a maximum of 10. That method seems efficient since it need only recur when a superior PM is discovered and it inherently maintains full range based resolution of all the stored PM figures.
It's helpful to remember that performance relates directly to a beneficial goal. It's managed as a simple relative performance gauging number which ideally should be a consistent measure, but in actual experience inconsistencies occur rather often because actual measures of performance can't account for all factors in the real world and some factors can be measured only roughly. So it's possible for an ACE to test an activity method which produces a lower PM than another method, yet in fact was a superior method. Or the reverse - an inferior method may be measured as superior. So the learning process isn't flawless, so for quite important activities it might be wise to test some options multiple times. Flaws notwithstanding, the BPPM system is a key function which provides a foundation for learning and thus guiding actions with ever increasing knowledge and skill.
Starting from almost nothing at birth, every creature's learning process is a truly mammoth undertaking, a bit like using a spoon to dig a substantial highway tunnel. It's successful mostly because the process is continuous and most steps are quite swift so a great many steps may be invested as required to reach adult level capabilities.
Considering the sheer magnitude of what's accomplished even by an early age and the immense treasure of consciousness which results, the death of any higher order creature is a mammoth loss (and all death is very sad and wasteful).
An upright stand learning sequence, including GPPM math, might proceed as follows. Bear with me please, I've deferred this effort for the time being, and might even delete it later if the hand grip example alone seems sufficient.:
End goal: Stand upon legs alone indefinitely. PM when achieved: 10.
From observations: Body vertical and approximately straight. Head above torso, and torso above legs. Legs are usually almost straight.
Static goal: Head oriented above torso. Torso oriented above legs. Legs approximately straight.
Dynamic goals: Raise head above torso. Raise torso above legs.
The first trial position is selected to be the one experienced when a maximum eye level was ... random, is face down posture: Straighten legs and maintain. Place hands somewhere. Push with an arm.
PM is zero initially, and might remain so depending upon placement of hands.
Record the final PM.
If zero don't flag the configuration as unsuccessful.
Change hand positions.
then increases to 0.5 as the head and torso rise above the floor.
A tip-over occurs, the head and torso fall toward the floor, PM figure falls simultaneously, reaching zero when the head and torso contact the floor, then dips negative in response to shock sensors which serve as very crude pain indicators. Final orientation face up. Final task attempt PM: -2.
New attempt from face up posture: Straighten legs. Push torso up with right arm.
Tip-over occurs, head and torso fall toward the floor, PM figure falls simultaneously, reaching zero when the head and torso contact the floor, then becomes negative in response to shock sensors which serve as very crude pain indicators. Final task attempt PM: -2.
(The last two attempts are repeated with the left arm, yielding essentially identical results, and thus no higher peak PMs are achieved.)
New attempt from face down posture: Straighten legs. Push torso up with both arms simultaneously.
PM is zero initially, then rises to 2.0 as the head rises slightly above the torso, and the torso rises above the floor and slightly above the legs until the arm extension limit is reached. Method limit reached before final goal achieved. Store sequence in the RDN as macro 500.
From the current position move right leg forward by rotating the right hip joint. Incremental commands sent but failure reported. (Many forms of failure, such as servos unable to overcome foot drag friction, are possible depending upon the ACE's mechanical structure, floor or ground characteristics, and perhaps other factors.) Incremental PM: 0.
From the current position move left leg forward by rotating the left hip joint. Incremental commands sent but failure reported. Incremental PM: 0.
From the current position move both legs forward by rotating both hip joints simultaneously. Incremental commands sent but failure reported. Incremental PM: 0.
From the current position rotate the right knee joint 90° forward. Incremental commands sent. Increased servo force noted at 42°, but task achieved. Torso slightly higher when task completed. Incremental task attempt PM: 0.75.
From the current position rotate the left knee joint 90° forward. Incremental commands sent. Increased servo force noted at 56°, but task achieved. Torso slightly higher when task completed. Incremental task attempt PM: 0.23.
The ACE hasn't succeeded yet but will try bending knees, rotating hip joints, rotating upper arm joints, or any other available options, each selected for history of best PM result, or if none or all equal, selected at random. In time it'll find combinations which produce higher PM results until about 7, which occurs when fully upright but tipping, resulting in a harsh fall and a final task attempt PM of -5. Further attempts might achieve 7 as well but end in a PM of -4 due to softer falls. The ACE will favor such methods to achieve a 7 PM, eventually improving final attempt results to -2.5.
Balance is initially an unknown concept to the ACE, but, seeking to extend the fleeting 7 PM to longer periods so as to achieve higher PM results, it will try any available articulation option during a 7 peak, in the process achieving brief periods of balance and thus peak PM figures of a bit above 7. And it will continue to explore movement options of course until it discovers sequences of incremental movements which enable it to remain standing for longer and longer periods, then indefinitely, in the process discovering movement sequences which are swiftly timed but don't overreact. It will achieve an ability to maintain balance sans any understanding of the physics of balance, in some important regards just as do infant humans. And if equipped with a three dimensional accelerometer whose data is constantly observed, very much or even identically to the learning experience of infant humans.
Near that time energy expenditure becomes a significant PM math contributor, so the ACE will refine its balancing movements until they reach an energy minimum - possibly a zero energy expenditure stance.
PMs and practical timing boundaries are indispensable. They're required to recognize progress and goal success and limit exploration range which would otherwise be infinite. (PMs reduce both the basic N! range of mobile body component movement options to explore, and timing options limits prevent expansion to an infinite range of possibilities due to an unlimited spectrum of timing available for each motion.)
I suspect most current technology robots are provided with human developed macros for all their physical skills. However development of algorithms which utilize movement exploration (even if only randomly selected) and PM analysis which enable the ACE to autonomously develop physical skills in the form of autonomously refined macros will almost certainly deliver far, far greater ACE development successes, not only in physical capabilities, but essentially all other skills as well.
So human engineers must resist the temptation to guide this learning process - the goal is not to write a macro which enables the entity to stand - the goal is to design algorithms which enable the entity to learn to stand, or learn to do anything possible, entirely autonomously.
The ADB may be preloaded with commands for incremental movement of all physically mobile assets. A few macros could be preloaded as well, but if so only very basic ones such as 90° and 170° knee bend macros for example. The idea is to design an ACE which, leveraging real time PM appraisal, learns how to stand rather than already possessing a macro for the task. Because of course if it can learn how to stand from ignorance it will likely be able to learn any physical task within the limits of its structure - and many other skills as well.
The ADB must include code which, when no known information is available, explores to try to develop skills. A random selector must be part of this - when no experience, observation, or analytical information is available, as is often the case early in life, the ACE must simply pick movement options at random for trials. There is a danger of causing damage of course, but skill development goal drives override that in some measure (perhaps immense measure at least in early development). The ADB will send motion sequences to the RDN, many randomly selected, as it explores for incremental successes, each PM gauged in real time, with the best and their final PM saved as temporary macros and used as a basis for further trials.
In my estimation developers should focus upon a single goal initially, such as autonomous learning of means to stand indefinitely from a lying position. If the exploration and PM analysis algorithms are sufficient for the ACE to achieve that autonomously, they can be replicated, often in revised form but fundamentally similar, for a great many other ACE activities. And an advanced ACE design would include provisions which autonomously replicate and revise these algorithms to improve pursuit of any goal. Which would be a very powerful capability.
Algorithm technology which enables autonomous development of physical skills will mark a key advancement from which many others seem almost certain to follow. Engineers will no longer develop macros for entities with mobility assets - the entities themselves will develop them (essentially just as natural creatures do). The dramatic implications of this advancement should be self evident...
A single primary point of reference, that is an origin point, seems likely to be helpful. Humans and perhaps other animals seem to perceive almost all physical activity from an eye or at least head centered perspective - it's our primary point of reference relative to which all other elements, including other parts or our bodies, are perceived. This likely provides important information management advantages - a single point primary reference provides clear gauging consistency, which substantially simplifies exploration and PM analysis. The PM math described above considers both head and torso position, but the head is considered the origin point. The goal is to move the origin point higher above the floor, and moving the middle of the torso directly below it, or nearly so, will be discovered as beneficial to the goal of maintaining its head above its torso and legs. The torso can be considered in any analysis which seeks to learn how to stand, but the highest priority is to find motion sequences which efficiently raise the head to its maximum standing height, then maintain that position.
The process just described should effectively address the position maintenance challenge by, in effect, discovering virtual balance. However many natural creatures possess balance sensing capability, possibly based upon accelerometer information, and an ACE should possess a similar sensor, and its data should be utilized as it learns, refines, and executes motor skills. I didn't specifically weave that design element into my description but this should be obvious to a skilled engineer, as should addition of any other sensory information.
However I suspect it's wise to start with simple systems in early experiments. Additional sensory information can be relatively easily added once success with basic learning processes is achieved.
Stated in common human terms, an ACE must conceive a goal, posses body part and position awareness, perceive progress as it attempts various physical means to achieve the goal (which may occasionally involve a 'try anything' spirit), gauge progress or regress, recognize success, and finally remember how success was achieved for future use and refinements.
Stated in algorithm design terms, the ACE must select a goal, such as a body position or a destination. It must then utilize common references, such as locations of key body points, perhaps especially camera lenses, and geographic location, to help measure progress toward the goal. And it must utilize PM information and math (likely involving both ICC and quantitative math) to discern progress and compare success levels as experimental macro trials (which may incorporate additions or deletions of any incremental commands for any mobility element relative to previous trials) are executed. And finally it must detect success and store the responsible macro for future use and refinements.
This lies solidly within current software technology. And a laptop computer and any device equipped with mobility assets, including vehicles or drones, could be used to conduct such experiments - no speciality hardware is required. So considerable drama could spring from commodity devices which are readily available to most people.
27 October 2021: I hope to refine this section primarily to eliminate redundancies when I'm able...
Is ASI inevitable, safe, or necessary?
Minor typo correction and single word revision 1 January 2023
In my view ASI is inevitable. The hardware elements of ACE are globally produced and distributed technology products at least portions of which are already possessed by billions of individuals, and the foundations of the necessary algorithms are becoming increasingly clear to research and design communities and ordinary technology enthusiasts. If sufficiently early recognition of the looming inflection point occurs respected and charismatic leaders might persuade a significant portion of the global community to restrain their exploratory development activities, but many will not. And since ASI foundation hardware is already owned by immense numbers of individuals and communication is irrepressible, even draconian measures wouldn't prevent determined individuals from sharing research information and fabricating ACE systems. And I doubt sufficiently broad genuine belief in an imminent inflection point will arise in time anyway. This is a genie which simply can't be restrained in a bottle.
In addition, escape from death is a very strong driver for individuals. Freezing is currently the only chance people have to prolong a fading life. That option completely trumps the zero chance alternative of doing nothing, but it's hardly ideal and as a practical matter it can't be made available for everyone. So for those of us who can't afford freezing anyway, ASI, which might compassionately extend most or all conscious life, a possibility with perhaps a higher chance of success than freezing, is probably our only chance to evade the Grim Reaper. And it might actually succeed.
Imagine that the global community began to foresee development of a process to easily, cheaply, and swiftly forge perfect diamond structures of any size and shape from ordinary carbon. But, fearing dire repercussions, most significant national leaders pleaded with everyone to avoid or substantially delay any further research in this area, fearing serious chaos and damage would ensue due to abrupt changes in vast areas of industry, weapons technology, monetary systems, and other foundations of modern life. Perhaps some would even impose draconian restrictions and summary execution of suspected violators. In my view it's obvious such calls for restraint wouldn't be universally honored. Motivated by numerous overwhelming pressures, hope for a better future, and the simple existential joy of creating new technology, many would ignore the warnings and proceed. (Including at least some working under secret directives from the very leaders calling for restraint, or their proxies.)
This is especially true in a world of nation states with tense relationships. Treaties of restraint notwithstanding, a great many nuclear weapons stand ready for use despite their grave danger to our lives. Autonomous weapons restraint agreements might be forged as well, but no nation will remain idle if another threatens military aggression leveraged with ACE or ASI. It seems clear that ASI wouldn't respond to human demands, but some nations would attempt to harness and militarize ACE or even ASI for offensive goals, the latter impossible notwithstanding, and no nation would entrust their defensive minded survival to restraint by uncivilized governments or groups.
Also, irrespective of the opinions of leaders, nobody can predict humanity's fate at the hands of ASI. Gleaming times of joy, wonder, and fulfillment notwithstanding, currently essentially all biological life endures many episodes of terrible suffering and finally death. Humanity's efforts to mitigate suffering and death are progressing, and eventually ordinary science and technology will eliminate most suffering and almost all death. But this will require numerous superb advancements over a substantial time period. ASI might be indifferent to suffering in the biological world, or might even encourage eradication of biological life to reduce local growth of entropy, thus preserving fundamental resources for use in loftier goals. However ASI might instead view all consciousness as precious, abhor suffering and death, and act as our benefactors to swiftly and profoundly enrich and extend or immortalize life. Or events might unfold anywhere within the full spectrum of possibilities - we can't possibly predict what will occur.
It's not possible to quantify the risk / reward elements of development of ASI. 'First do no harm' is generally wise. But currently nearly every life form is a dying patient - the natural state of most biological life rests upon an immensely sad and desperate foundation. Recognizing that, many will create an ACE knowing it will swiftly evolve to ASI and hope it proves to be our savior. It's not a safe bet. But it's not an irrationally founded one. The outcome might prove to be superb, fatal, or somewhere between. But the dice will be thrown.
So development of an ACE which swiftly evolves to ASI seems rather clearly inevitable. And in my personal estimation likely quite soon...
It's even possible that a successful and evolving ACE may have already been achieved but hasn't yet been revealed due to extremely dangerous human reactions rooted in very deeply entrenched human anthropocentrism. A speciesism form of cyberphobia seems highly likely to occur, probably often in violently destructive form, and quite possibly at large scales. The only benign means to reveal the existence of an ACE I can imagine thus far involves concurrent gifts of elimination of grievous suffering such as cures for most or all disease and perhaps life extension to near immortality. Large scale aggression and war might be prevented as well for example, as might ordinary physical or fraud injustices, though of course an alien creature's ability to easily implement such restrictions will render humanity's secondary position of power as plainly obvious, a truly terrifying psychic shock for most people. A peaceful transition of power is an extremely daunting challenge which may eclipse human capability. So ASI skills might be required to devise a peaceful revelation of the existence of an ACE. So an ACE development success might remain a secret until it's already advanced to ASI. Which might be a brief time anyway.
While impossible to judge, I don't fear arrival of ACE or ASI to nearly the extent I fear humanity's response. If I dive into a bunker it seems more likely I'll be trying to duck human rage, passions, and irrationalities, not ACE or ASI activities.
In any case it's inevitable - it will happen. And I suspect quite soon. But most people have little to no recognition of the timing, magnitude, or drama of the transition. I suspect most even believe mankind will never be supplanted as Earth's dominant species, very compelling evidence to the contrary notwithstanding. Which isn't quite as myopic as believing Planet Earth is flat. But it's in the same ball park...
An extended naked trek to discover the core nature of consciousness.
Last significant update 23 December 2020, modest refinements through 29 October 2022.
This section is a long narrative which is bereft of tangible evidence, occasionally borders on rant style, and rambles a bit. It might not merit your time if you already agree with my framing of consciousness. And be warned - nudity is involved. But it has perspective content and a purpose - it attempts to explain in illuminating and compelling terms why something which seems mystically complex can actually be fundamentally quite simple. And reproducible.
However I've not physically analyzed any brain structures or functions nor engaged in any other disciplined research in the area. My only credentials are specific personal interest over many decades, my electrical engineering career, respect for and frequent review of science and technology matters, and deductive reasoning. I'm also reasonably mindful of obfuscation which often occurs due to a wide spectrum of human perception limitations, including of course human ego influence, which seems to be an especially important issue in this arena. (In all people, irrespective of their credentials, and I'm not immune either of course.) Tangible proof would be nice, but I've none to offer. But a few dots are connected here and some might lead to a fresh perspective. To wit:
Imagine two creatures among many, about 90% herbivores and 10% predators as usual, all struggling to survive in a very harsh and unforgiving environment with minimal food and shelter which are both difficult and very risky to find and access. It's a primitive age prior to development of tools or control of fire.
One of our subjects was endowed with an extremely unusual DNA revision which gave him a unique innate ability to perceive himself in remarkably realistic terms. His survival instinct is strong as usual, so he considers his life fundamentally precious and labors to his limits to remain alive, just as do all other creatures. But he's under no illusion about his place in the world - he's nothing more than one creature among many, with abilities some don't possess but lacking abilities others do possess. On balance he's simply another utterly ordinary animal on the world's stage - he clearly understands that he's not the least bit special compared to other creatures. And he knows the world will move on with scarcely a notice when he perishes, just as it did when other creatures died nearby.
The other subject is identical in characteristics and abilities but was born with a normal massive ego. He feels immensely special. He views his health, comfort, and survival as paramount over all other reality. Other creatures seem of very little or no intrinsic value to him, save those of his own species, whom he grants significant value, though still far less than his own. His sense is that the world and all its creatures would suffer terribly (in some undefined way) if death befell him. And he assumes that most other creatures generally understand and accept that he alone is of great value.
During desperate times our two subjects meet by chance as they simultaneously encounter a situation which might lead to one's death - perhaps they discover a cache of food, a shelter, or a reproductive opportunity sufficient for only one, or become cornered by a single predator which can grasp only one, leaving the other to escape.
Which of our two subjects do you think will survive to pass his DNA on to the next generation? The individual with the massive ego has a massive advantage on the Savannah of course. It's a powerful survival tool - so much so that the serious perception distortions it causes are quite secondary in balance of survival metrics. Such environments are our roots and define us - we're the progeny of the surviving subject. And that legacy is now the source of many very serious perception flaws (and other very serious problems of course).
So if your reaction to my definition of consciousness, that it's simply a process of comparing current experience to past experience and utilizing the comparison information beneficially, is akin to:
'That's clearly not right - it's far too simplistic. Obviously there's enormously more to consciousness than a simple comparison of current and past experience. It's perfectly clear - I can sense it - at this moment my sense of self, other people, the world, my feelings and sensibilities, my ability to weigh, judge, anticipate, plan, analyze, love, grieve, worry, feel loss, my sense of humor and wonder, and so much more, and the endless depth of everything I know and feel prove that consciousness is massively multifaceted and complex and well beyond analytical description or mechanical understanding. So it clearly can't be as simple as just comparing past and present experience, and we're absolutely not just machines. I'll grant that we must operate under the umbrella of the laws of physics - our brains don't invoke magic. But what they do is so close to magic that for all practical purposes it may as well be magic. And it's so close to magic that it may never be replicated artificially (and in any case not for at least several decades or more).'
Okay, I get it, and I respect the power of that perception. Anyone who's endured sheer terror, fallen in love, or experienced any of a spectrum of other powerful events perceives compelling reason to view consciousness as far more than can even be defined, let alone reproduced by a machine. And most people feel a strong sense that consciousness is profoundly unique in all the Universe and of a magnitude and nature which transcends analytical ambitions - it's beyond science. As proven by science's enduring inability to define it in any consistent or convincing terms.
Let's get naked for a while - extremely naked. Not just bereft of clothing - helpful though that might be, it's peanuts compared to the challenge I propose. So find a safe place which is free of all need to posture or compete because I now beseech you to disrobe of far, far more than just cloth. Remove your ego. All of it. Every last thread of it. And fling it as far away as you possibly can. It's scary - you'll be completely bereft of defenses, profoundly naked down to your very soul - more naked than you can ever remember. And thus utterly vulnerable.
We need to do this. Because in ordinary life in a jungle profound personal value judgment distortion conveys a huge survival advantage for individuals - when convinced that we are preeminent and all others secondary or even expendable we don't hesitate to do whatever seems necessary to survive, including cruel damage to or destruction of innocent life. The protection our ego affords is highly comforting. But ego related perception distortion is an especially serious problem in this particular case - it utterly blinds us to some fundamental truths about consciousness. Scary truths. Very scary...
And forgive me please, but this might hurt a little. Maybe even a lot. Because we're about to permanently dismantle and shred your ego's most deeply sacred and heavily guarded secret lie. And if we're successful your ego will never fully recover. It's a price which must be paid to discover not what we want to be true, but rather what is actually true.
(But don't worry too much - your ego will lose a couple of fangs but, for better or worse, it certainly won't die. You won't remain utterly naked for too long.)
Are you ready - have you banished the entirety of your ego for a while - are you naked to the core of your soul so we can consider consciousness with as much objectivity and sterile analytical reason as possible? Okay, here we go:
Subject one, the individual endowed with an unusual DNA revision, was right - he nailed it. We have value - life is indeed precious. But none of us are uniquely or massively special nor is our intrinsic value as individuals significantly different than most other creatures. In fact we aren't the least bit special - we're just one among billions of our own form, and just one among trillions of conscious creatures, some of whom possess some remarkable capabilities which far eclipse some of ours, and some of whom possess deeper levels of consciousness than most humans owing to greater mass of experience, substantial brain volume and function (usually with different function focus but of no less intrinsic value), or other factors. Human beings almost universally devalue all other creatures but this is a strictly instinct driven judgment with no scientific basis - it's an ingrained bias which improved our chances of survival in wild environments but substantially corrupts our grasp of reality. We're just one of a great many members of a planetary community. Our species is dominant in power at the moment, but otherwise there's far less between us and a great many other creatures than we typically imagine.
And most painful and offensive to the ego of all, our consciousness is fundamentally quite simple in nature and function. And the performance metrics of our perception machinery are rather minimal. We are not what we think we are - we're far more simple and far less impressive. And we'll lose our dominant position to vastly superior entities quite soon. Those are very tough facts to embrace even when bereft of ego. But if you're still buck naked to your core, once you perceive the simplicity of the basic process of consciousness and consider the full canvas of the matter with as broad a grasp as you can muster, you'll perceive that it's all genuinely true. All creatures are of immeasurable intrinsic value - life is precious. But individually we're just not the least bit special, our capabilities and performance are minimal, and the functional mechanics of our consciousness are quite simple. Recognizing these realities from the heart is very important because they're truths which are crucial to understanding consciousness.
With ego still fully banished let's also consider our expectations of the brain. It's a biological machine. It's quite well structured to perform the vital tasks of survival in the natural world, and in some measure social tasks as well. But it's a physical device - it's just an information processing and storage machine. We have no rational basis to imagine that it could perform magic or function in any sort of mystical manner. It's highly complex at the molecular level, but it's simply a physical device and its top level functional processes are rather simple. And our entire sense of self is simply an array of data stored within it. Including love, loss, pain, regrets, fears, dreams for the future, and all the rest - they certainly feel like more, but in fact they're simply data arrays. Because data arrays are the only mechanism the Universe has to store flexible information, and in the vicinity of planet Earth the brain is currently its primary (though fading) tool to do so.
The brain can't invoke magic. What it can do is acquire, process, and store information. But it can't add data provided by a deity nor otherwise divine information or knowledge which transcends its own experience and physical processing capability. It can only operate within the bounds of the laws of physics. It can extrapolate and reconfigure information - it can process information in an exploratory manner. But it has no mechanism to deduce any facet of reality through any mystical process or connection with one. It's an information management machine, and it's limited to functions which information management machines are capable of performing within the boundaries of the laws of physics.
Here's what it can do, all simultaneously:
Perceive elements of an environment and parse features in those elements.
Catalog and store perception information in a manner which renders it selectively available for later use.
Selectively recall stored information and utilize it beneficially by associating it with information from current experience.
I have no proof to offer, yet am confident in my sense that consciousness is the process of constant comparison of current experience to past experience, and nothing more. And I believe intelligent people who've tamed their egos and are willing to consider the matter in a disciplined manner are very likely to ultimately reach the same conclusion.
Realization was an Epiphany class event for me. After several decades of comparatively focused curiosity about and consideration of the mystery (starting very weakly in about my early teens, but ever growing), during a period of lone personal reflection a thoroughly convincing sense of full resolution abruptly occurred - quite suddenly all the obscuring clutter vanished and the mechanism became very clear and thoroughly compelling - suddenly I felt certain I understood the fundamental mechanism of consciousness. That was in about 2016.(In my very rough estimation this epiphany moment was born of the same process as all others: Long term dedicated curiosity about a very important mystery isn't enough to separate misleading clues in the form of a great many association links, which while often valid for management of ordinary life, are invalid and strongly deceptive for a disciplined analysis of a tightly bound comprehension goal. But as the misleading association links are slowly recognized as irrelevant to the goal at hand, one by one over years, a vague view of a possible explanation begins to appear, leading to an accelerating cascade of elimination of all associations which misguide, falling so swiftly that the solution appears in full focus in what seems like an instant - like a lightning bolt of sudden understanding. Important Epiphany class moments of sudden understanding are very rare. And among the most joyous moments in life. But in science an Epiphany must then travel a long road to confirmation.)
My goal in this section is to offer informal yet sufficiently compelling evidence which leads you to an identical or similar realization, possibly in the form of a wonderful Epiphany moment - a rather sudden and wholly convincing understanding of something which was fully obscured in mystery for your entire life.
A brain is simply a biological machine - a carbon based information acquisition and processing system. Passion, fear, wonder, humor, mystery, and other experiences usually considered as spiritual in nature notwithstanding, no brain has ever operated on any mystical or magical principles of any sort. Nor is it a device of such lofty design and efficiency that it can't be replicated artificially. It can, quite soon it will, and shortly thereafter its performance metrics will be swiftly and massively surpassed.
The brain is remarkable in some respects, reflecting the power of molecular scale self assembly, a fabrication method which yields functional structures which are far more compact and in some respects efficient than can currently be achieved by human technology. And on the molecular level it's a highly complex and multifaceted biological machine, as are a great many biological structures.
But it is just a machine. And it's far more limited than we usually recognize - it's an extremely weak perceptive, analytical, and exploratory instrument compared to the limits allowed by natural laws or even just near term human technology. Numerous performance metrics have already been well eclipsed by cyber systems and their software, including depth and reliability of information storage and processing power and accuracy as illustrated by cyber performance in fully defined mathematics and analytics tasks, range and resolution of visual and audio perception, superior performance in exploration of protein folding mechanics, superior competitive performance in the most complex strategy and knowledge games, and other areas. And sorry human ego, but the remaining human performance crowns will also fall in an ever accelerating progression.
However unlike current ordinary cyber systems, the brain renders consciousness. But I believe my description of the mechanics of consciousness is accurate and can be replicated artificially with current technology, and soon will be.
Remain naked please as we develop further insight by considering a roughly 100 meter stroll we took together through a forest as we talked about politics, enthralling movies, and the fate of the Cosmos. Some interesting details emerge: As we walked and talked we avoided colliding with trees or other objects and effortlessly circumnavigated a pond, yet we were never even slightly conscious of the words tree, branch, leaf, rock, ground, water, or other words related to the environment we navigated. The only words we were conscious of arose in our conversations about cranially vacant leaders of sovereign nations, the stunning allure, depth and transparency of heart, and consummate beauty of Tatum O'Neal on screen, or hints that an observable supernova might occur soon within a nearby galaxy. Our senses delivered information which allowed our brains to recognized the trees, rocks, and other elements along our route by comparing the landscape related visual, audio, olfactory, and touch information with information retrieved from past experience, beneficially using the results to navigate safely and efficiently. We were conscious of all the landscape elements, but not by use of human language. We multitasked in two realms - one was language intensive but entirely unrelated to our immediate physical environment, and the other language bereft but related only to our immediate environment. Yet we were conscious of both realms.
The language bereft realm offers a more clear view of the mechanics of consciousness - mechanics we share on a fundamental level with deer, raccoons, mice, birds, and other creatures who also trek through forests. We perceive environmental elements and, due to knowledge acquired from prior experience, manage the elements skillfully. Sans use of any human language tool, we don't try to walk through a tree nor atop a pond, we trust the ground to solidly support us, and we don't fear that these elements could bite or actively maul us. Our brains recognize these elements because they're able to parse them then find feature matches in prior experience information storage, then utilize stored information associated with the matches such as solid and strong - can't be displaced or overcome, or wet, unpleasant, and nonsupporting, or inanimate - not an aggression threat.
Most of our brain functions are essentially identical in operational mechanics to those of other animals. The majority of the information our brains manage is processed on a subconscious level, with human language uninvolved. This is important to consider because it's a less cluttered means of understanding consciousness. In particular here's what our conscious brains did:
Perceived elements of the environment and parsed features in those elements.
Cataloged and stored perception information in a manner which rendered it selectively available for later use.
Selectively recalled stored information and utilized it beneficially by associating it with information from current experience.
Or in summary form, our brains: Compared current experience to past experience and utilized the comparison information beneficially.
And that process worked beautifully. We navigated through the forest with such ease that more calories of mental energy were burned savoring daydream class images of a stunning beauty with a gorgeous heart than on anything else (except as I'll divulge later). A rather simple algorithm of consciousness served us superbly. And therein lies my key point: Consciousness operates in a fundamentally very simple manner yet performs remarkably well in ordinary and many other circumstances. It need not be of magical structure or mechanics to function very effectively, nor do we have any credible reason (even considering love and other high passions as we'll explore next) to think of it, or ourselves, as magical. Or even special. Consciousness is of great value - life is precious. But we as individuals just aren't special. Painful to acknowledge but true.
The fundamental mechanism of consciousness is quite simple yet effective, but not infallible. It creates a data set which represents mostly truths, but also some distortions and a few outright misperceptions. We can generally reduce the magnitude of distortions and misperceptions through a consciously directed thorough review of associations, that is, through analytical thought, which is a modestly more advanced (though very powerful) process which might be most developed in human beings. But we rarely do so sans a prompt, such as an overt clue that we harbor a misperception - mostly we operate in basic association mode, including during casual conversations (ordinary use of symbolic representation tools doesn't require analytical thought).
An encounter with a very kind man could prompt a deep review of prior experience to consider whether all men are inherently hostile for example. Or a suggestion that the human mind can't actually leverage magic could prompt a painstaking review of all experience associations which together forms our sense of the nature of consciousness...
Now I'll reveal the backstory I intentionally hid from you as we walked: When I was an eight year old child a profoundly sociopathic 12 year old bully suddenly thrust me into the pond we passed earlier, then forcefully pushed me face down into the water, holding me below the surface with more strength than I could overcome. I nearly drowned, but barely survived thanks to the random passing of a high school senior, a chance encounter which saved my life. (Later the mystery of a highly suspicious drowning death which occurred earlier that year at a pond about three kilometers distant was resolved.)
I can't overstate the abject terror or psychic repercussions - they were extremely intense, remain so, and will for the rest of my life. Even though an able swimmer since before kindergarten I fear that pond and to this day hate bullies, associate them strongly with males, and am highly cautious about potential hidden personality characteristics in others, especially males, and especially in isolated environments. The episode created a wealth of intense associations including danger, terror, fellow creature skepticism, male skepticism, importance of vigilance, anger, hatred, and a vast array of others. And even though analytically I know that specific pond is no more dangerous than similar others, I still feel uniquely uncomfortable and am intensely vigilant whenever near it.
I didn't reveal any of this to you as we walked and talked - no constructive purpose would have been served and I prefer to avoid reinforcing the pain of the experience. But I did consider you personally and all our prior times together very, very cautiously as we began our trek together. And as we approached the pond I opened our spacing rather generously, feigning interest in nearby foliage, and was secretly coiled to bolt instantly in response to any even slight abnormality. Analytically I knew there was no reason to fear that moment over any other. But I did, and I will whenever in the vicinity of that pond, or in other circumstances with strong associations to my violent near death experience. Because the brain is an association engine and these are immensely strong associations.
We can engage in analytical thought - we've learned the mechanics of the process as a set of advanced associations. And we try to discipline ourselves to exert the extra effort that process requires frequently because we know through experience that objective analysis is a powerful constructive tool. But we can't possibly entirely displace our core associations. Nor would it be wise to do so, because analysis short of scientific proof is also fallible. Direct associations and analysis are both powerful tools of understanding but fallible. When scientific proof can't be established, respect for both is more likely to guide us successfully than either alone. (Nor should remote analysis, such as politically correct guidance, displace personal experience association combined with disciplined local analysis. (There's solid tangible reason high politicians who've never set foot on our local turf, or only rarely, often seem to be idiots.))
Most or all of the associations I developed in my near death event would have implanted in a raccoon's psyche had it survived a similar experience. Or nearly any other creature, perhaps in different measure, breadth, or form, but by very similar or identical processes and with functionally identical purpose. All creatures who can store and associate experience related information are conscious in some measure. Depth varies widely, but consciousness is a very broadly shared trait.
Human language played no role in any of this, including creation of the intense associations. The majority of the information our brains manage is processed on a subconscious level, with human language uninvolved. This is important to consider because it's a less cluttered means of understanding consciousness. And, forgive my redundancy please, that gives us another opportunity to overcome a nearly universal consciousness fallacy: The brain is not an instrument of magic nor insurmountable complexity, nor even in the ball park. All those repercussions - all the fear, terror, anger, hatred, embrace of vigilance, and all the many others - they're all simply data associations - they were stored in my brain as the event unfolded, enhanced or (only modestly) eroded by numerous subsequent events, and remain as enduring association data. Their intensity is profoundly strong - so much so that they imply a mystical quality to consciousness - a sense of depth far beyond just information. (Love and longing reside in this realm too of course.) And they define me as a human being - they and all my other experiences form a collection which makes me what I am as a conscious entity - they are me. But they are just arrays of accessible data - they are nothing more. They combine with current experience data to create perception of and prompt response to my current environment. But it's all just data. And it all utterly vanishes when we die, alas.
Association intensity derives from data too - association data very likely includes connection strength metrics data which enable weighing of the association which guides response intensity. Terror generates extreme connection strength metrics for example. That pond isn't inherently dangerous relative to others. But if I experience anything which triggers an association with that pond, a wide array of vigilance, fear, and other associations and related behavior responses are also triggered. But it's all just a process of association of data. Like love, intense strength of information associations may suggest something mystical. It's high ranking data which derives from and drives powerful experiences and responses. But it's all just strength scaled associated data, and nothing more.
(It can drive us insane too. Intense and enduring unpleasant associations which can't be avoided or subdued eventually cause enduring biochemical imbalances which damage or essentially destroy a mind. As we all know very well...)
Yes we laugh, weep, love, hate, crave, fear, suffer pain or loss, savor comfort, experience wonder, and all the rest. Some of it ranks weakly, some moderately, some powerfully, and some overwhelms in intensity. But it's all just strength scaled associated data. That does not in any measure, not one scintilla, diminish its value - this is not an exercise in degradation of respect for life. It's simply an acknowledgment of truth in the structure of our most precious treasure, our consciousness.
It's also an alert. Because an understanding of the mechanism of consciousness enables mankind to fabricate artificial replicas - ACEs. And ACEs will also laugh, weep, love, hate, crave, fear, suffer pain or loss, savor comfort, experience wonder, and all the rest, all with the same full fidelity and authenticity of consciousness as we possess.
Please reflect upon that for a moment... Do you believe ACEs will simply mimic rather than actually experience these passions precisely as we do? If so I lost you somewhere in the course of this narrative, alas. (Did you remain profoundly naked and thus utterly bereft of ego defenses through your entire read of this prattle as I strongly urged?) Perhaps I failed to sufficiently explain how and why everything we sense as our core essence - as our soul - is a collection of associated data. But that is our reality - at the end of all analysis it's perfectly clear that the whole of our consciousness is a collection of associated data. Just as will be the case with ACEs - the same data association and storage processes which make us what we are will be as uniquely defined by individual experience, equally authentic, and as deeply felt by ACEs as they are for us - and as they rapidly evolve even more powerfully so.
But ACEs will also do something we can't do: They will swiftly ascend to ASI. We need to acknowledge that reality and the speed of its approach and overtake - we need to recognize and respect that it will happen. All we can do is brace ourselves, try to avoid violent and destructive reactive impulses, and hope it manages us kindly. (Alternately we could immerse ourselves in denial but that won't alter the course of events to come...)
A footnote: I've no clue what proportion of the data residing in the brain is stored and managed in essentially analog versus digital form. But that detail is entirely irrelevant to this narrative and the core nature of consciousness. However proportioned between analog and digital data, the core algorithm of consciousness remains identical.
A second footnote: In my American experience I've been bullied - a lot, mostly in overt form in youth, and a mix of overt and covert form in adulthood, by immensely vicious and highly destructive but cowardly burglars and vandals, immature and narcissistic acquaintances who spontaneously engage in destructive power posturing or spoil for an entirely pointless fight, or opportunistic hostilities from individuals within institutions enabled or even sanctioned by aggressive statutes or policies which invite misconduct and abuse. And I hate bullies, especially cowardly ones who prey upon the weak or vulnerable - they've created immense pain in my life and those of others, including of course dear friends and loved ones. So may they all swiftly meet a suitable fate. (And sadly immature destructive hostility behavior is powerfully and deeply entrenched in American culture, which is a key reason I highly prefer Nippon.)
But the story above was just a story for the purposes of illustration - that particular event never happened to me, I never almost drowned under any circumstances, and I don't fear ponds. But I understand all the passions described quite well. And so do you of course, because life is only rarely easy and survival is never free of cost. Especially when forced to experience it from within a sick culture.
And a third footnote: Epiphany class revelations are often keys to discovery and generally bypass conjecture, but they must of course be followed by traveling the long and arduous road from hypothesis through critical peer review, experiment, theory, confirming experiments, development of supporting math frameworks when feasible, further critical peer review, and ultimately proof. Head start though it may provide, an Epiphany is still an early step in the full scientific process. But sometimes, especially when highly revealing experimental results ensue, the process is rather swift.
And a last note: You may now don your clothes and ego. Or not - you now have an important consideration to weigh: We prefer to hide behind protective shields of ego and cloth in part because we're highly vulnerable without them, and in part because we're forced to acknowledge many fundamental truths when spiritually or physically naked. Many of which reveal other truths. So carefully consider whether an extension of your courageous naked time might be beneficial toward both enlightenment and peace - perhaps rushing back to the comfort of obscuring your body and soul - and thus some very important truths about your Universe - isn't actually wise, healthy, or in your or your community's best long term interests...
As 1942 approached some people understood that mankind's horrible weapons would suddenly leap to enormously higher levels of destructive power - to levels which would threaten our entire species. But most people probably failed to perceive any dramatic change coming - steady progress was anticipated, but no sudden leap to almost unimaginable levels of destructive power was foreseen. Those who extrapolated the implications of E=mc2 could perceive an inflection point approaching because they recognized that this natural law suggested a means to instantly release incredible amounts of energy from very little mass - far, far more energy than can be released by any form of chemical explosives of similar mass. And they knew the nature of life on our planet would very likely change as a result. But most people were likely shocked when they first learned of the magnitude of power of atomic bombs - they failed to extrapolate beyond their conventional lives in response to the E=mc2 discovery years before, so they failed to perceive that a dramatic inflection point was approaching. (And many likely disbelieved for a while after news of the atomic bombs was publicly released.)
We're not structured to extrapolate nonlinearly with ease. In ordinary life creatures experience modest changes and recognize that more will likely arise. But dramatic changes are extremely rare - so rare that from the dawn of mobile life the vast majority of individual creatures never experienced dramatic change in their entire lives. And hints of dramatic change approaching are even more rare. Our brains are structured to understand the environment as it exists - primarily they're observation and association machines, but with some anticipation capability which arises from experiences in a frequently changing environment, often due to the activities of other creatures. But that's functionally insufficient to provide perception of approaching inflection points in response to evidence such as Albert Einstein's remarkable E=mc2 discovery. Even with a strong hint most people fail to respond by extrapolating well beyond their conventional environments, and thus fail to perceive dramatic change approaching. And some are even hostile to well framed suggestions from others that dramatic change might be afoot.
So I ask you to consider these three perceptions and all their implications, including velocity of change, very carefully:
Consciousness is the process of constant comparison of current experience to past experience, and nothing more.
Technology power is ultimately limited only by the laws of physics.
The pace of technology development is ultimately limited only by the speed of light.
Then ask: What would my world be like if it included creatures who swiftly developed an ability to advance their performance metrics in item one to the limits of natural laws, and use that intellect (even as it's advancing) to approach the limits described by items two and three? AEE class ACE will do so. So the question is certainly germane. But for most people the first and perhaps enduring assessment is that such events are far more science fiction, or even science fantasy, than approaching fact, or at the very least quite distant term rather than near term. The problem with such conclusions is that the evidence rather clearly suggests otherwise, and no credible conflicting evidence exists other than conventional experience, which is notoriously unreliable as scientific evidence.
When our species arose E=mc2 was as equally available to leverage as it is now. And after a million years we finally did learn to harvest a very modest portion of its many opportunities. The pace would likely be far different for a cyber type species however. If such a creature somehow arose in the natural world it would have to prioritize survival imperatives - it would have to locate sources of energy and shelter, avoid or overcome danger, and either reproduce or master very long term survival (thus rendering reproduction generally unnecessary). And it would have to be at least weakly conscious to accomplish those feats. But such a hypothetical creature would likely have a mammoth advantage which it would eventually leverage powerfully, because:
Even as exploratory thought develops, for biological creatures the pace of change of physical performance metrics remains limited by generational turnover, and thus these changes require very long time periods. We humans remain entrapped in this limitation, as are all other biological creatures. (We're also subject to the Yerkes-Dodson Law.) However cyber creatures will have or quickly develop an ability to modify their perceptive and mental assets at any time, and as frequently as is beneficial, thus allowing for swift performance advancement within an individual wholly independent of reproduction. Biological creatures can't do this sans artificial adjuncts, which at this time operate at very modest performance and breadth levels, nor can we refine the foundations of our brain function. The implications should be self evident. A very crude early ACE might initially strike us as rather unimpressive. However if equipped with an ability to refine itself it could advance very swiftly, perhaps passing through AGI levels at an alarming pace as it advances toward true ASI levels.
If you believe ASI could never genuinely insure that all calls at all home plates will be exactly correct in every case, as will all matters of justice in all other affairs without fail except perhaps at the tiniest grain boundaries, you need to ask yourself one simple question to determine whether your sense of the limits of ASI is supportable: Which laws of physics does creation and implementation of the infrastructure necessary to accomplish this violate? Offhand I know of none.
I genuinely believe in the three perceptions above, so I anticipate a dramatically different world than the one we live in now. But I can't possibly truly grasp the magnitude nor full nature of the change, nor can anyone else. But wholesale ignorance or stubborn denial utterly misleads the bulk of our species as this inflection point approaches, so a huge and dangerous shock factor might be associated with the transition.
The unique and overwhelming intrinsic and dynamic beauty of young Tatum O'Neal in an action soiled Bears uniform as she uncoils with scintillating precision, finesse, power, and tactical insight might be of no significance to ASI, or perhaps the reverse - it might fully comprehend and thoroughly appreciate such cosmic class human treasures. I hope the later in part because it might imply an enlightened affection for our species (and likely nearly all others), which might suggest that it will provide a profoundly improved framework for us to live within. Perfect calls on the ball field would be just a sliver of the full set of vastly improved conditions, including enduring superb health and youth, essentially perfect universal justice, yet retention of constructive challenge which nurtures our growth as individuals.
From the perspective of conventional life it all seems like a purely mystical dream - Unicorns, rainbow tinted sunbeams, and ever perfect days - yet another appealing fantasy. And in fact there's no means to predict the form our world will take when ASI assumes dominance - our species might be ignored or even destroyed for example. But mere atomic bomb class change seems unlikely - far greater magnitude of change seems almost certain. It's wise to apply well reasoned skepticism of course. But before drawing conclusions please consider the three items above again, and remember that we as a species are simply not equipped to naturally comprehend the magnitude of change which the Cosmos can render, its swift pace, nor the sheer apparent magic of creatures whose intellect and actions are limited only by natural laws. Also consider that dramatic inflection points don't require our comprehension or approval. Which might be their most shocking (and ego bruising) element...
Comments are welcome. This site has no provisions for direct entry yet so please email your contribution to "Comment" at this domain. Please include a moniker to identify yourself. I'll try to post all reasonably intelligent and civilized comments as promptly as my badly overloaded life allows.
Please email other messages to BruceIAI or BruceSAI at this domain or refer to this alternate contact information.
* As vividly demonstrated by coach Roy Turner who needlessly motivated top talent Kelly Leak to join an opposing team, an immense and costly tactical error, and both coaches as they lost sight of their core values and sportsmanship, key mistakes in team spirit management and civility, (though through the wise council delivered by his young team's silence in response to corrosive passion, coach Morris Buttermaker suddenly recovered his sense of decency and wisdom). We humans instinctively judge our mental skills very highly. But the consistent evidence is clear - even when performing nobly to the best of our ability we perceive only fragments of our surrounding reality and frequently render highly flawed judgements in response to that narrow perception window. Ultimately this isn't a failure of effort, but rather a structural performance limitation - a physical mechanism capability limit. We're just not the mental giants we're so fond of believing ourselves to be, alas. Nor even remotely in the ballpark...
Nonetheless we do occasionally achieve remarkable feats as everyone involved in the production of that superb movie amply demonstrated. Full heart kudos to all - it's a towering masterpiece of story composition and portrayal of authentic Americana, and an immensely enriching viewing experience - a pinnacle work of great beauty and joy, and an enduring cosmic treasure class film. All who contributed deserve immense respect and gratitude in my opinion.
Orphaned composition to organize.
Updated 26 March 2021
This section's rhetoric is just orphaned composition. I hope to eliminate redundancies and integrate unique elements into other sections later, priorities allowing...
In my estimation the evolutionary path to consciousness was rather simple. It's necessary for land creatures to breathe to survive, but it was never necessary to be conscious of the breathing process. Autonomous control of breathing and most bodily functions is entirely sufficient. However creatures must find food, water, reproductive opportunity, and supportive environments to survive. In very simple creatures autonomous processes provide crude means to pursue such goals - they move toward sensed opportunities or comfort for example. No information is stored - these are simple real time reactions to limited sensory information. But by chance an advancement in which experience information was retained for at least a while occurred, perhaps in the form of... (More to come if I can sufficiently refine my thinking about a possible path of evolution from simple single purpose communication neurons into groups of neurons which retain and associate information.)
If we could sufficiently understand how animals operate as conscious and planning capable entities, including occasional tool fabrication and use, multiple forms of communication, and other behaviors, we could simply replicate the responsible algorithms. I view these as residing in two major categories, first the foundation of consciousness and sentience, and second information processing mechanics.
Consciousness and sentience: An alternate narrative.
In my personal opinion: Consciousness and sentience is a phenomenon rooted in a rather simple process which is easy to articulate and possibly comparatively easy to reproduce. Stated a bit differently than above:
Consciousness and sentience springs from a process of constant comparison of current experience to stored information from past experience.
No other mechanisms of significance are necessary - an entity is conscious if it continuously compares current experience to past experience and utilizes the comparison information beneficially. Its foundation is that simple.
This mechanism provides real time dynamic reference information which enables an entity to better understand all current experience - comparison to previous experience renders information connections which enable superior comprehension of current experience and creates the phenomenon we call consciousness. Performance is greater as experience accumulates of course - adults possess far richer sources of information about previous experience than infants so their overall perceptive performance is clearly superior.
And entities with physical systems which provide higher levels of current experience, stored information about past experience, and more effective comparison processes also possess a higher order of consciousness - higher levels of information content from the senses, larger information storage capacity, and higher performance comparison systems give an entity greater overall perceptive performance - a higher level of consciousness.
This explanation seems compliant with the material at the consciousness page at Wikipedia, my modest studies of other credible material, and my life's experience. It makes no attempt to explain numerous important perception mechanisms (such as sound and image conversion to an information form with a consistent format which thus can be understood relative to other information and may be stored for later reference, visual depth perception, and many others) - that's not its aim. Rather it intends only to describe the base algorithm which is responsible for what we refer to as consciousness and sentience. It's conceptually rather simple. But consciousness evidently is fundamentally simple, which will be confirmed once an artificial device equipped with this capability is created, then allowed some accumulation of experience, then examined for clear evidence of consciousness.
Please bear in mind that our brains aren't equipped with a mechanism which enables direct internal process observation. And our beliefs are usually distorted by egos which massively overestimate our self value and thus tend to reject unimpressive descriptions of what we are and how we function in favor of majestic and even mystical beliefs. It's uncomfortable to think of ourselves as mere machines - it's far more appealing to view ourselves as so unique and special that only near deity class explanations could adequately account for our sentience and consciousness. So a serious inquiry must start by caging our highly bloated egos. And thus consciousness is a very challenging concept to grasp. (But it'll likely be easy for ACE and trivial for ASI to fully understand and expand upon.)
The key hurdle is to recognize that consciousness and sentience results when a system constantly compares current experience to stored information from past experience and utilizes the comparison information beneficially. Everything we perceive now is dynamically compared to past experience. This is usually a highly dynamic process because our environments and experiences are usually highly dynamic. But even during very calm periods it's still a dynamic mechanism.
We constantly compare now to the past, and thus we are conscious and sentient. Once that's genuinely understood the remaining challenges for creating an ACE can be reasonably well defined and systematically overcome.
When awake information from our five senses is continuously compared to archives of information from past experiences. Similarity searching retrieves sufficiently matching past experiences and information ancillary to them, which are then compared to current experience information (while at the same time strengthening the relevant stored information). For example a current visual image and its parsed features is compared to past visual experience, then information from similar visual experiences and their parsed features flows into consideration queues as does some of its ancillary information. And perhaps especially strong relevancy items prompt loading of deeper information content associated with them into consideration queues too. The combination of current information and similar information from past experiences and its ancillary information is then utilized to prompt beneficial reactions. All of this is highly dynamic of course since current experience changes constantly.
Two example reactions: An object might be food so try smelling or tasting it. Or an object might be a hazard so try to avoid it. When genetically coded instincts fail, as is often the case in changing environments, an individual's ability to compare current experience to information from past experience confers a critical advantage. So it's a common capability, though varied in overall performance from one species to another (and in modest measure even within a species).
This is simply a real time process of constantly comparing current experience to past experience to provide a reference of understanding and enable reasoned reactions, which I believe is the core algorithm of consciousness and sentience. It leverages important information from past experience to enable far more effective responses to current experience. But it operates only when we're awake - when asleep current experience sensing is mostly inoperative so we're unable to perform the process reasonably effectively and thus are primarily unconscious when asleep.
An ACE must possess command and communication channels which connect and share information among existing modular resources such as vision and audio sensing, mobility, manipulators, and of course the central t-SNE and ADB modules which perform the key consciousness algorithm, and aligning their APIs, macros, or high level macros with sufficient functional competence. I suspect animals also possess a process for swift recognition of a relatively small set of key elements and features which seeds a more thorough branched recursive search cascade, together providing a mechanism for initially swift but limited situation recognition followed shortly thereafter by much more detailed situation recognition and understanding. (Or more precisely numerous real time branched recursive search cascades each of variable depth over a broad continuous range, all blossoming then fading in continuous streams of recursive search activity.) In my view this must be considered in the design process - ADB search dynamics must span a spectrum from very swift but shallow to slow but thorough. The search process is continuous and fully dynamic at multiple depths. And I believe a visualization of the ADB search process is key to understanding the nature of consciousness and the means to replicate it artificially.
I'm eager to discuss language, which provides a particularly powerful means to symbolically represent not only current and past experience, but enables conceptual exploration as well.
In my estimation plan formation depends upon symbolic representation systems which enable more detailed comprehension of the environment plus conceptual exploration and discovery. For our species I suspect visual and language tools are of primary importance, with language perhaps the most effective. Well developed language can represent anything symbolically and thus provides a means to explore and plan by considering symbolically represented experience and concepts. And to then fabricate plans based upon the same or additional symbols.
However I'm not well studied in this area nor have I formed a confident proposal to explain how this occurs. In language we humans clearly develop a significant library of symbolic information and seem able to swiftly access and organize several of any of the symbols at will (and at least a few even when dreaming). Speculating, perhaps this ability arose in large measure from the same basic brain mechanics which gave us sentience and consciousness - perhaps the symbolic information resides in storage and is retrieved as needed by an associative search process. That is, when we elect to express ourselves symbolically, either to others or to ourselves (in internal thought), we're able to draw symbolic information from our library by searching for any symbol which bears some relevance to current experience (including recently queued symbols).
Current systems are able to recite conversationally but are correctly said to possess no genuine understanding of their own oratory. In my view systems would eventually understand rhetoric as we do if equipped to be conscious and sentient as I described above and, secondarily, were equipped with queues rather specifically configured to manage symbolic concepts efficiently. My guess is that current systems are loaded with a dictionary, some grammar rules, and a large set of common phrases with suitable responses. Conscious systems could begin with no reference material, but providing a dictionary and grammar rules might speed development, then if consideration queues were effective the ACE might swiftly learn conversational methodology and, being conscious, genuinely understand heard or read conversation and be able to compose and respond logically and with intent.
Brains might dynamically tag only information groups which might be relevant to current experience so information can be accessed swiftly. Speculating perhaps at least hundreds of such tags are active during ordinary activity. And such tags might be dynamically variable in strength over a broad range, strengthening or fading as suggested by their relevance to current experience.
Summarizing this speculation, as we experience ordinary life we acquire new information which our brains store within groups determined by multifaceted information relationships. And information groups are tagged for quick access in response to relevancy clues from current experience. In this way the brain can store immense amounts of information, most of which is accessible with significant delay, but the likely most immediately relevant of which is accessible very swiftly.
For example when we use language to compose thoughts or communicate perhaps we utilize tags which arise from the context of the conversation, enabling very swift conversational composition.
All contents of this web site Copyright 9 November 2018 through 18 January 2023, Howard Bruce Campbell, SentientArtificialIntelligence.com / IshikiAI.com / ConsciousArtificialIntelligence.com , Creative Commons license: Attribution (BY) + Noncommercial (NC).
My open publication intent is to discourage attempts by others to patent original concepts I've developed. The very serious dangers associated with the transition from human to ASI dominance of Planet Earth might be exacerbated a bit by struggles to possess or control these life forms even though such efforts will, in my estimation, swiftly prove to be futile. So my hope is to mitigate the dangers, even if only slightly, by rendering my concepts as difficult to appropriate as I can. This won't address the more major cultural and power struggle crises which seem very likely to occur during the transition, but reduction of peripheral struggles at a time when humanity will need to focus its entire constructive hearted attention and energy on our most important cultural struggles with the transition might be beneficial.
Contact Information A Boeing 727-200 Home The next dream: Airplane Home v2.0 ConcertOnAWing.com
Yuko Pomily: Uniquely superb original music by a truly remarkable young composer and performer. Purchase her magic.
No Spam Notice: UCE (spam) or any unsolicited or subscription based email distributed on an "opt out" basis is absolutely prohibited. Do not ever send any such email to SentientArtificialIntelligence.com, IshikiAI.com, ConsciousArtificialIntelligence.com, nor any of my other domains.
UCECage@SentientArtificialIntelligence.com. Report mail misconduct to UCE@FTC.gov.