(see one of my last, previous posts for the perspective on Psychology)(not really much translation involved, in the following):

The aspects of my perspective that are typically missing in AGI's (artificial general intelligence) ideas of "cumulative learning" [(see "Cumulative Learning", By Kristinn Thorrison et al )] include:

True Hierarchical learning -- which is more than a new category and a responses using and building on existing behaviors PATTERNS of otherwise of the same nature (as those used in the past).  (Note the thinking always in PATTERNS -- it is thus that behavior patterns show detectable changes, which are the new behavior pattern itself AND by which behavior patterns are defined (as in classical ethology) -- i.e. DEFINED BY THE PATTERNING OF BEHAVIOR "SURROUNDING" THEM.)  Then realize: True hierarchical learning ADDS new elements (in the behavior/response pattern), which shifts some key pattern(s) _AND_ which promote (IS) fuller sensing/perception of the key aspects of the situation(s) BY the Subject -- literally PERCEIVING NEW CONCRETE ELEMENT(s) (as indicated, AND these may be across several times and across several circumstances -- especially later in ontogeny; such is the power of SOME of our Memory systems; we simply must use "more imagination" here, both the Subjects and as researchers/theorists).  The sub-elements (lower level elements OF the previous responses to "such" situations) may change in their nature as they are used differently (e.g. "tagged" or "typed"),  or at least when associated with new-different circumstances; some may certainly be truncated or dropped out (think: new "chunking").

Foundations in PERCEPTION -- yes, THAT kind-of basic process.   It is with/in perception (and later, attention FOLLOWING THAT) that  provides for (IS) "new elements which promote full sensing/perception of the key aspects of the situation BY the Subject literally PERCEIVING NEW CONCRETE ELEMENTS [(or elements in a distinctly new context)]".   Resolving that seeing SUCH new things and JUST THAT (see above), as the foundation of each new level of abstracting ability (i.e. abstraction) -- THAT is a major seemingly paradoxical set of "things" which simply must be resolved ("bucking" the philosophies of the past).

Ontogeny involves a new type of learning at each stage, unfolding in response to (or included in the response to) NEW elements of the concrete situations/circumstances  (and, given the sophistication of some of our Memories: this can be across times and spaces.)  Here, it is important to see/find TRUE ANALOGIES (not just "trumped up" analogies).   These are doubtlessly useful in generalization to "other" circumstances -- seeing other situations similarly better by seeing MORE there "too".  PLUS: We must get rid of the idea that "learning" is always the same type of thing IN ACTION; it changes qualitatively there, BY VIRTUE OF CONTENT, AND CHANGING RESPONSES TO THAT.  Ironically, in my system , in another sense, all learning is the same in that it conforms to simple associative learning patterns-- that is all that is needed (or likely), given what else is going on.  [ Of course, good integration, consolidation and generalization of earlier behavior patterns must occur before "moving on" from one stage-type ("level"-type) to the next. ]

Thus, the AGI machine must contingently, after previous developments and integration/consolidations/generalizations, SEE MORE) BASIC [(here meaning: additional)] ELEMENTS OF THE SITUATION.  And, JUST THIS, provides for moving in-key-part(s) the whole system -- allowing more abstractions (things seen conventionally as "more abstract"), and THUS yielding more refined responses (whether they are specialized or not -- to some extent an open question -- BUT THEY ARE NEW w/r to the important sets of overt, express, explicit circumstances (AT LEAST clear at the inception of such a new sort of processing)).  Likewise the BEHAVIOR PATTERNS, AT LEAST AT FIRST ARE ALSO overt, directly observable and clearly expressed. It is important to realize that although initially overt, directly observable and expressly and explicitly seen IN patterns of behavior, such overt-ness of direct, observable overt evidence of change may be short-lived, as the Memories change and incorporate the new (new type) of learning behavior (perhaps VERY quickly) (This is why, for humans, eye-tracking technology and associated technology (e.g analysis software) likely have to be used.)

Given the distinct limitations of short-term memory (I should say "working memory") and the LACK of limitations of other Memories (e.g. visual-spacial) make it understandable that small changes in response (including PERCEIVING) must be able to yield BIG changes in understanding; this is why this perspective and theory make sense (and ONLY something like it could make sense).  AGI simply must figure out such ontogeny as I have described AND DO IT.  In AI you have the great ability of trial-and-error, quickly and over-and-over, that allows for a fair amount of guessing (I would guess) -- and give the "locality" of the beginning of new patterns in behavior COULD (in theory, with a thoroughly educated view/approach) BE GUESSED AT.   But none of this is possible without an appreciation for True Hierarchical Learning during ontogeny -- very, very likely occurring in qualitatively different stages.  The machine must make ITS OWN analogies, and only such analogies are appropriate (as has been the case in science "forever" ).

Something very much like I propose (above) OR attempts at AGI (as is and has been the case with Psychology) can continue-on, basically the same way as they have been for decades -- i.e. no big progress (as is acknowledged, again and again in the AGI field).