Hello Brad, thank you, when you asked me - in January... I realy did not know how to answer...) Now may be I also not very strong) but I can start to... We pressed the button "START" - and AI-entity had been born... Thè kernel of AI-OS should start... The pure technical service processes - are out of review, - we need to clarify some substances... How i see it just now. 1) What should we start from: space-time continuum - the only initial one. My answer - time. This is the first "identifiable-and-definable components". Of course, the "child-AI" do not need initially dd.mm.yyyy hh.ss..... - it realy does not know what to do with it... Yes, we could "clone" robots (AI-agents) in very simple way... and keep "the minds" of "previous generations" - we will do it in some way in sure - we have to decrease costs of the thraining, - but initially I will start from "0" - the current state just now) Time as the concept of course will be partially "identifiable" - inherited - but not in complete manner: initially - from the "start" it will be the "order relation" of the signals from the sencors: "before-later-at the same time"; then - it concerns the some characters-parameters of the signals .... later it will concern the concepts identifications and pattern recognition events ... This is the only I can express just now. Thank you for your question. I will see thoroughly your works later - the work overload prevents to do it just now) I am very interested to exchange the minds with you. Kind regards, Sergii

Dear Sergey Korneev  

That was a good, interesting over-view of the beginning processes of your AI machine.  Perhaps my writings might help you "pin down" the pattern recognition and concept identification aspects.  But, your description sounds good .  [ I guess the most interesting aspect of my view might be that all major types of "patterns", no matter how abstract, have some NEW concrete (real, in-the-world) elements (or type of element) at there inception; specifically, you might like to read my stuff to come to understand and see something about the nature of the type of new key part of each new level of concept, and with another key type of element as you go from one level of conception (concepts) to  the next, and then similarly to the next ... (up to 5). (Concepts, of course, are hierarchical and memory, including episodic memory, and declarative, procedural, etc., brings forward the abilities and developments (learnings) associated with the last level of concept as part of what comprises the next level of concept.)  This might give you clues as to what and when key new aspects (of recognition) should enter your system (and perhaps help you with ideas on their nature and the nature of their eventual products -- concepts and related abilities). ]

Thanks for sharing.

Regards, Brad Jesness

Article A Human Ethogram: Its Scientific Acceptability and Importanc...

Dear Brad, thank you. I will see it in details and come back. As I told before - I am concentraiting now on the first if not seconds but hours-days of AI-entity living. You put the core issue of this development - "identifiable-and-definable components"!) The challenge is:  more "identifiable" - more rigid system but costs of training - less; more "definable" - more flexible and universal system but costs of training - more. In every task/objective of AI-robot the optimal balance will be other. Optimization of the costs from my point of view needs "cloning" of "AI-agents" till they are adecuate to the environment - AI-agent can change many "bodies" and keep the knowleges and rules inherited. When we decided that AI-agent became very hard - we need "to clean" its memory and start the process of "new training" - mostly "selftraining". Thank you for your mind. Best regards, Sergii 

Brad, I started with my AI-individual definition - namely - "The Definition of Individual Intelligence. The Individual Intelligence has been defined by self-controlled “border” (in physical sense)... " when I was observing my children had been born. It was clear for me that when child was born the only point belonged to him - the point between the eyes a a little bit above. From this point child started "to draw" its body - "the border of Intelligence". We should make some to prevent the child to wound the eyes - because of initially child do not know that eyes belong to him. - From this observation I started. In 1983... ) Sergii

Dear Sergey Korneev  

Of course the most serious limitation to what my view can offer you is that the stage* shifts (yielding new ways of conceptualizing), which I strongly argue can be just perceptual (perceptual/attentional) "shifts" , have NOT yet been discovered.  They should most certainly be discoverable with the new eye-tracking technology (and computer assisted analysis); but, until these discoveries, they are in many senses not a sure thing.  [ I did try, based on empirical findings seen by major classic theorists, try to define some nature to these 'shifts' (or at least, in summary: associated observable effects) which I hope will be helpful. ]

I am not in the position to do this research, so we all will just have to wait and see.  I look forward to the hypotheses being proven or disproven.  (Either a psychologist or someone in another field who studies cognition and development should be able to do this research. )

Regards, Brad

* FOOTNOTE: The shifts in ways of seeing yielding qualitative changes may or may not be rather abrupt; there may be some very notable common thread(s) to them.  Thus, it is possible, for those who highly question 'stages' to see the developments as more continuous, and not-so-much stages (but, in any case, there are qualitative changes in conceptualization abilities -- and an empiricist finds some important observable concrete bases for them, though they may be subtle).  Fortunately, from my perspective, neither the continuous/discontinuous debate or the nature/nurture debate are really relevant any longer.  Neither of these old 'debates' keep us from moving forward (they no  longer need to be settled a priori and we need not have certain dualisms at all, in this basic context).  This is helpful.

P.S. Dear Sergey Korneev 

To put it in the least words (one long sentence): The saving grace for seeing and understanding developmental psychology and for advances in other fields (like AI) as well is: the great likelihood that each major level of abstraction (conceptualization) has some clear concrete (in-the-environment, observable) aspect(s) and they will not be hard to see as patterns, because:  given their context of development (the context that previous development provides), they are likely rather simple (specifically: perceptual shifts). 

Today, with eye-tracking technology we can find them, if they exist (and I believe it is very likely they do).  Already we can somewhat know their nature by clearly patterned associated observable effects (empirical observations already existent, related to existing historic and major theories -- described in "A Human Ethogram ... "), but the ultimate, essential nature (and necessary "trigger") for each successive level of cognitive development may very well just be perceptual (perceptual/attentional) shifts:  This is BECAUSE (to put what I said before in other words) , very conceivably, this is all the more required in an adaptive complex of behavior, including the significant memory capabilities that carry the key representations of experience forward. 

There is NO evidence against this position and it is more biologically-likely than positions taken by other modern modelers and theorists.  Other approaches do not work for explanation, understanding, or for good research.  All the concepts, language and terms, of classical ethology can be used if one starts from this ethogram theory perspective and necessarily applicable biological principles can be adhered to.  The ethogram theory position is as empirical as possible (and provable/disprovable) and will exclude no good existing findings.

Dear Brad, thank you, I will think it over...)

Dear Sergey Korneev 

I hope you give it a lot of thought.  You do not really want to impell your robot along the course of development with just the key behaviors it starts out with nor (as an alternative) do you want to have the robot just trained by people or imitating them. 

For inspiration: Think of the largely fixed patterns in universal language development  or think about true guilt and how it reliably appears only about 3-4 yr. of age; think of the routinized courtship displays of some birds which only occurs with adulthood; think of the calls of birds, which no matter how notably stereotyped, come only with development and does require learning (and otherwise do not appear in some species if the birds are isolated). With such thoughts you might see how it can/could be very good to keep innate factors AND learning and development together.

Paraphrasing Lorenz (from some text):  "This means that to predict behavior in natural conditions it is necessary to know what the animal’s innate perceptual and behavioral instructions are (as in Uexküll). In the same spirit, he claimed that without the notion of innate blueprint it would be impossible to study learning (Lorenz, 1965; Lorenz in Schaffner, 1955, p. 144). His argument is that stimulus association needs a releaser to which a conditional stimulus can be associated, and that random response variation alone is improbable because learning almost always results in adaptedness."

Hello, Brad,

Please, see my definition of meaning "Thought" in Artificial Intelligence - for your overview.

Definition of the concept “Thought” in Artificial Intelligence (by Sergii Kornieiev).

“The Thoughts can be of the following types:

-          Thought Type 1 – “Imagined “scenario” that was formed by: objects, concepts, and activities (virtual) “of them” and “with them”.

-          Thought Type 2 – “Analysis of properties/characters and their relations – “in static” or “for period time”.

-          Thought Type 3 – “Concepts/object definitions”.

-          Thought Type 4 – “Classification/Pattern recognition”.

-          Thought Type 5 – “Reasoning/Conclusion”.

-          Thought Type 6 – “Decision “to do” the real activities according to the Thought Type 1 in the State 2 (see below)”.

Every Thought Type can be in one of two states:

-          Thought Dynamic State – in active software process.

-          Thought Static State – thought properties/data had been stored in the memory after the active program process had been completed.

The Thought Type 1 in the Thought Static State can be implemented as the “motivation to do” and “scenario/plan/schedule” of real activities of AI-robot with the correspondent decision – the Thought Type 6 in the Thought Static State.

Dear Sergey Korneev

I have given my responses to the system you outline elsewhere and will repeat them here, since they are so pertinent to this thread:

I am quite supportive of the view (outline) in your very short paper. I have only 2 things to add: (1) You must come up with a way of understanding all the qualitatively different sorts of abstract conceptual thoughts; the KEY is, it is widely agreed, they are hierarchical and use main aspects of earlier levels (types) of abstract thought as components [in some way] in later "more advanced" levels of abstract thought. And, (2) I believe we must come to see what is added from the environment (input) to the end-stage earlier thought, that triggers (and is, in a sense) the inception of the next stage -- by empirically finding major subtle but overt perceptual/attentional shifts and biases (likely using eye-tracking technology and sensibly programmed computer-assisted analysis).

The various types of memory and their natures and roles is another important matter (what has to be represented and in what form, with what accuracy, ETC.). Then, emotions -- not that hard a matter, since these really are very basic patterns -- yet functionally noteworthy. And, then: PERHAPS language OR, more specifically, tips we can get from language about the nuances of thought-processing. (Language aids thought, but it is NOT thought to be core to coming to have different types/levels of conceptual (abstract) understanding or thought; and, even given its toughness to USE, one should still not fail to find any usefulness it has in providing key clues ("tips") about very specific aspects of processing and how they vary with development. This should not be overlooked -- especially since this may be both a more basic matter and an easier type of matter,)

Perhaps the same main outline you have of the types of "seeing"/processing WILL DO, but it will have to at points shift ITSELF, accommodating to the shifts in their conceptual subject matter's 'abstraction' level AT ALL POINTS AND in ALL WAYS of processing (in/of ALL the major sorts you outline): specifically at 5 different progressive conceptual thought levels. I believe if we can learn enough about humans (esp. the major perceptual/attentional shifts), then we shall have the knowledge to know when and how to have qualitative transitions.

Your outline, in any case, is a good start and thank you very much for sharing that.

P.S.: The NATURE and qualitative difference of the 5 hierarchical levels of conceptual thought is roughly indicated in my longest paper, "A Human Ethogram ...". (The associated paper on "Information-Processing Theories ..." is also worthwhile. And, no doubt, the 100s of pages of my essays on researchgate could be of some help.)

P.P.S. I would also advise that you keep the "steps" in your processing as situation-driven (or situations-driven) as possible (so it is more like one big process, with phases OR elements). Quite amazing things should be possible within the capabilities using the very substantial capacities of the types of memory. In other words: avoid the merely mechanistic: this would be analogous to having an [artificial] homunculus ("man-within-the-man"), something many psychological models are guilty of. Behavior should flow and transition as it really does in human phenomenological reality.

Article A Human Ethogram: Its Scientific Acceptability and Importanc...

Thank you, Brad, the "hierarchical levels of conceptual thought" is out of my understanding yet... I will think it over.... Thank you for your deep article! I will see.  ... "avoid the merely mechanistic" - it is hard to follow when you need to program AI...) but I will try Regards, Sergii

Dear Sergey Korneev 

The hierarchical system:  You just have to decide when your Subject (in your case a machine) has developed ("learned") enough, actually conceptualized enough, at the first level of perception/action/memory/representation that it could use (to great benefit) applications of some of THAT, at least some of those select entire systemS (groups of overall formulated knowledge) as units in a higher level of conceptualization (_YES_, there are more than one system at each level ; AND:  the units, which themselves ARE the previous systems, may be in some sense "trimmed").  IN short, essentially your  outline of the process (which you can already program) can be used at EACH level, but with different units (AGAIN the units at each level are themselves systems -- systems, in a very, very real sense, even during in the first level of development).

The precise nature of what is described (outlined)  IS NOT KNOWN, but roughly the nature of these systems and units are indicated in my paper, "A Human Ethogram ...", enough for you to see the systems _and_ systems of systems as the hierarchy they are (hypothetically) and as qualitatively different (each, a REAL SHIFT from earlier thought).  Knowledge of the real nature of the systems AND the "perceptual/attentional shifts" that are the inception of the next level requires longitudinal, developmental research with humans, very likely requiring eye-tracking and computer-assisted analysis technologies (and this research may be done by AI people and/or psychologists).

But the programming at each level can use YOUR basic system, just using different units at each level, and somehow knowing when (and seeing how) to transition. 

By the way earlier levels of conceptualization continue to operate, as appropriate, even as more "abstract" systems of thinking are developed. ALSO: not all previous systems will progress to the next level.  Also, you should recall, there are a total of 5 conceptual levels (aka, levels of "abstraction"), all roughly hypothesized and describe in "A Human Ethogram ...".  (Unfortuanately, this may not be a simple case where if you can do one level-change in your programming, the next will be easy -- it surely won't be easy, but perhaps may be easier. Likely significantly easier would be to do a "proof-of-concept", having your machine progress through all 5 levels in a limited and clearer "problem space".):

I believe FIRST and most important you should try to approximate this entire hierarchical conceptual development system and make it work in a program.  Understanding the various types of memory and their nature and roles is a very  important matter (what has to be represented and in what form, with what accuracy, ETC.)  These memories are considered CAPACITIES, basically (but perhaps not completely) similar in their natures at each level (obvious the content or content accessed differs).  Only then worry about emotions and what clues can be abstracted from language usage. You do not care as much what may be a joy or threat to your subject (the machine) as you may for a human, but it is true you might well need to push those things which are an interest or which are surprising.  You only really need language, if your computer needs to communicate with humans (this is a VERY complex add-on to human abilities, and of course aids some learnings, etc.; maybe someone else can do (or has done) this and you can use it as an add-on too -- but you will have to set up the interface as well (on all appropriate levels).

[ MAKE SURE YOU READ THE MATERIAL BELOW THE FOLLOWING LINE: ]

_____________________________________________________________

GENERAL FOOTNOTE:

If you have to or if you want to view the levels in a mechanistic way, you could see each subsequent level as involving, in some way, a larger collections of conditions (including some KEY new environmental aspects, or the salience thereof) AND where that collection is VERY notably important _AND_ yet earlier levels are still also very important and thus identifiable or distinguishable from other levels.  Perhaps such mechanisms may still essentially have the natures of the levels.  NOTE: we cannot fail to consider "aspects of the environment" that exist only by virtue of a temporal/spacial relationship or difference (anything one of the various memories can "keep track" of may be the "aspect of the environment").

There is no doubt some gain in efficiency and/or other advantages to adaptation** having the levels of conceptualization, so your system should also display this (in my human examples, in my big pager, this shows).

** section FOOTNOTE: adaptation may include "needed for adaptation, given limitations" (but no limitations should be presumed, they should be seen or necessarily so).

Article A Human Ethogram: Its Scientific Acceptability and Importanc...

Thank you, Brad, I have read it and will return to it later - I cannot be sure that ALL you mentioned I understand clear but in general yes...) I can say just now about the definition of ''Thought" proposed: "Thought Dynamic State – ..." - this will be more complex, in sure: the Thought can be stored but not completed... and also some other kind of hierarchy - dependencies, impacts, some other .... - I am not ready to formulate it just now - may be in communication with you it will move more fast...))) Kind regards, Sergii

Dear Sergey Korneev  

In some correspondence (Messages) some have been concerned with getting emotions properly into AI, without "borrowing" the models or theories of psychology directly.  I will address this matter below :

Outside of major study on the various memory capacities and their natures and capabilities (which is THE most important thing to do, along with creating a machine that develops types of concepts/behaviors and thought in a stage-like hierarchical matter -- previously addressed (above)) :  the other thing which people (esp. some psychologists) often think is very important is emotions.   Let me address the matter of emotions for AI, here now:

Do not make too much of emotions (as some psychologists certainly do). If you can make hierarchical learning mechanistic AND devise proper storage capacities (representations/internal models (memory capacities and abilities)) [ which also determine the "environment" (circumstance(s) or set of circumstances) responded to ], which are the most important things, then adding in basic emotions (OR THEIR FUNCTIONAL EQUIVALENTS) will not be something very hard. Emotions are highly patterned: meaning that they are clearly triggered by a TYPE of circumstance* (NOT one circumstance or situation, in particular) AND the responses which they are IN THEMSELVES are very simple until put into cognitive terms (which everything ends up in): it just amounts to orienting to the trigger and providing a certain general sort of motivation ("energy" or activation) and THEN everything a rational being (or machine) does after that will be largely a sort of cognitive response (OF THE SORTS PREVIOUSLY DISCUSSED) appropriately matching the circumstance: With the cognition, you have a matching of the circumstance CLEARLY BETTER THAT THE EMOTION ITSELF DOES.

It is possible to view emotions as cognitive helpers (for us poor creatures, just as they are for other mammals), helping to prompt an appropriate TYPE of orientation and TYPE of response. It is conceivable that these could be mimicked , adding on types of orienting and types of motivation, BY adding such properties to THOUGHT processes. (IN short, as long as what the emotions DO gets DONE, you, in effect, will have provided the guidance aid which emotions are.)

Certainly do not get "lost in the woods" of emotion (which to say it another way: it is a superficial sort of response, by itself (that is, without cognition)).  I recognize emotion as important, but it is not a huge problem and gives you no reason to get off your basic, present AI track.

You also do NOT have to mimic the full range of emotions (all the emotions) and certainly not all the emotional STATES, IF your machine has a way to examine all possible alternatives in a challenging circumstance and pick the best one (perhaps being much more effective at that than a human -- needing less "help"). On the other had the machine does need to know if it is really in any way confused or lacks information.

Certainly many of  the nuances of human life DO NOT HAVE TO BE MATCHED -- there are simpler and more efficient functional equivalents. (Also emotion responses often have to do mainly with problems in social relations; the relevance of these are related to how much you are going to have your machine have social relations (if just for practical work, the emotions will surely be a more restricted set and I doubt you are going to have your machines seek every sort of intimate relation).

* FOOTNOTE: The fact that emotions may come into play in "very complex" circumstances is NOT due to a feature of the emotion(s) itself.  The "very complex" stuff is a function of the 'seeing' of the environment, via perception and cognition and memories -- THAT is the setup.  The emotion is still triggered by just the basic type of the circumstance.  NOTE: This is NOT to say that the experience of the emotion(s) may not be qualitatively different (perhaps partly due to an admixture of emotions) NOR (correspondingly) does it say that the emotion would not get a different label (this might well account for at least many of the various STATES of emotions -- for example, see: http://atlasofemotions.org/#states:anger ).  This perspective KEEPS YOU FROM thinking you may have to individually program many different emotional states.

A personal P.S. : The reason why I am taking such an interest in AI is that I am not ruling out the possibility that AI will SHOW in their modeling of human behavior a more impressive ability to predict behavioral patterns and sequences and actual realistic behavior changes under circumstances, than the models of psychology itself. (And THAT will really show -- since robots do ACT (actually instantiating operational definitions!)).