AI is bad at general intelligence. Each AI app has to be separately coded to do what it does, we do not yet know how to code an artificial general intelligence that can pick up new skills. For example a medical diagnostic app cannot drive an autonomous vehicle, while a human medical doctor probably does hold a driving license. Even understanding a regional accent is beyond many voice apps - for example some perform particularly badly with Scots and Newcastle (Geordie) accents.

AI is also bad in unexpected situations because it is only as good as the programmer's foresight. For example an autonomous guard bot, designed for flat ground, recently failed to spot some steps down into a pond and fell in.

Machine learning has a long way to go, not only in generic capabilities for the autonomous learning of unanticipated tasks but also for the identification and definition of those tasks.

And of course, if you want a full model of human behaviour then such things as emotional responses, ethical beliefs, personal preferences and so on all need to be included, along with the ability to intellectualize about them all. We do not even know the depth of our own ignorance here, but some model of the self will probably be necessary. Would properly human-like behaviour involve consciousness? We have no idea how to code for that!

Dear Guy Inchbald  

Thanks for your interesting and cogent response.  You certainly seem very familiar with AI.

Very partly realizing some of the challenges you cite, I always try to think in terms of offering an outline of all human cognitive behaviors which, while sketchy, could function in all the multi-dimensional aspects in any defined and very limited setting; basically, it seems to me that such a demonstration of proof-of-concept could be quite impressive at least for those who know what they are seeing demonstrated  (even while the actual behavior range is outrageously limited).

The outline of human behavior (which I have already tried to provide) is totally empirical (each aspect) and everything (each and all capacities and capabilities) are empirically related.  Thus it is fit for computer coding and programming.  [ ( I have left the full sorting out of all the aspects of the human I have described to the engineers themselves -- since the sorting out needed should be clear to one who is capable and they need to process it for themselves anyway. It may well be best that I am not involved in the details -- though I stand by the system as an integrated wholly empirical system. *   I also did not even bother to describe the roles of the simpler systems at work (e.g. emotions) BECAUSE good information on their roles and nature of their aspects is already "out there".) ]

In short, I believe I have done what I can and I hope it is all the basic input from psychology needed for one (an AI engineer) to make a proof-of-concept demonstration.

I believe most of the problems that seem hopelessly challenging basically only seem that way because people want to do everything at once -- when one "bits off more than they can chew", things tend to be confused and confusing. .  For a very meaningful proof-of-concept, you need to do much less and just close enough to "right" that it is fixable -- just get everything (using everything in the model) working well in the right direction, in as limited a setting as necessary so a  demonstration of the workings of everything can be seen.

[ To refer you to references to my system, see the posts under the Question:

https://www.researchgate.net/post/How_would_an_AI_robot_with_all_useful_human_abilities_and_human_capacities_differ_from_a_real_human_and_how_need_it_not_differ? Concepts you do not find in the description of my basic system OR at least mentioned (e.g. emotions) are either secondary (e.g. motivation) OR not needed (or false or wrong) **.]

*FOOTNOTE: To get to my empirical system (with all major elements central to all cognitive developments and learnings), I did have to challenge 6-7 basic "assumptions" (actually presumptions, because they are unfounded and unproven) of mainstream psychology; these presumptions are/were destructive of real and true and likely possibilities (plus the key hypotheses involved in my system are TESTABLE (verifiable, replicable)); these presumptions, beyond being limiting, are/were MISLEADING and destructive to having as much of an empirical foundation and embedded relationship with the environment which is otherwise, and LIKELY AND INDEED possible (EXAMPLE of that which is misleading in modern, mainstream psychology:  having developments of "abstract abilities" occurring entirely in the mind -- with no clear environmental referents).  The alternative assumptions I use are more likely and more biologically related than the "assumptions" they replace. [ My system needs no "meta-" capabilities or special separate "executive functions ...".  THUS, my system solves the hopeless man-within-the-man problem (i.e. the homunculus)  (which could never be resolved); the capacities and capabilities/abilities properly (mutually) balance themselves within a good system -- the concept of homeostasis in some way comes in here and there (something not found in any real way yet in modern mainstream psychology).  I could mention other advantages of my system (solving age-old debates) which are relevant to psychology, but I am sick of addressing those stubborn ideologues, and totally unresponsive "boneheads" -- and you need not hear of all that HERE.  Psychology cannot argue for itself, as is; and, psychology cannot argue against my position -- they have done nothing even close in almost a year. ]

** FOOTNOTE: My perspective on human development is the only system offering BOTH a top-down view of behavioral development and a bottom-up view.  The mainstream views are basically irrationally bottom-up, i.e. JUST "learning";  I cover both learnings AND realistically also cover maturation due to innate guidance factors involved at key points WITH learning (something the mainstream really doesn't cover at all in express terms of real distinct behavior change).   BOTH a view of learning AND of maturation factors (involved intimately with learnings) are necessary for reasonable distinctions being made between TYPES of learnings (otherwise: learning maintains just a mythical as well as a very vague and unrealistically, unlikely ubiquitous status as a sole mechanism).

 Dear Dibakar Pal  

I do question whether AI machines could be better overall than a human; but they could be better in a lot of ways, while self-learning/developing; they would be different, yet the using the same systems -- but this need not imply that anything would/could get "out of control" (nor do I think this is likely -- as we learn more about the human, we will have a good sympathetic/empathetic response, I believe).

AI engineers (or perhaps I should say: computer scientists) could AGAIN (as in the '80s) provide a model for psychology; the psychology field itself is worse than stagnant, worse than "in a rut"; it is irrationally bounded by unproven assumptions, this yielding truly ridiculous "theories" (for example: see the "embeddedness" 'theories', which can/could only provide weak and extremely indirect 'evidence' -- so much so it indicates not that the 'theory' is correct, but that it is wrong!).

 100 years ago, people say computer is bad, because it try to replace human brain, they paranoid about human no longer needed to think by using computer...today, everyone using big or small computers, be it smartphone, ipad, ipod, laptop, servers...same apply to AI/BI/Bigdata analysis, it is just another tool for human to use for improvement and can produce better results in production, business, research, study, medicine, education.. it is not going to make human useless or brainless, it will not replace human and make human jobless, but it will help people to focus more thinking/mental power on higher priority or more important things rather than less important, repeatable, monotone activities and tasks...

Dear All 

AI needs my perspective and the elements and relationships I outline (my theory). Let me indicate how and why this is so.

[ I put the most of the following in a FOOTNOTE just added to another post, but believe it deserves to be seen, so I make the statements HERE as well, with some additions, so all this will be seen. ]

My perspective on human development is the only system offering BOTH a likely and realistic top-down view of behavioral development and a bottom-up view.  The mainstream views are basically irrationally bottom-up, i.e. JUST "learning";  I cover both learnings AND realistically also cover maturation as it (in my clear hypotheses)  affects learning (in real time) due to innate guidance factors, INNATE ACTION PATTERNS, operating WITH key learning -- of course, all in response to specifics in the environment (perceived and attended to).  Maturation is something the mainstream psychology really doesn't cover at all in express terms of real distinct direct behavior change -- I DO.   BOTH a view of learnings AND of maturation (at key times operating essentially a part of the important learnings) are necessary for reasonable distinctions being made between TYPES of learnings (otherwise: "learning" maintains just a mythical status, as well as a very vague status and is seen unrealistically as a sole mechanism and having a ubiquitous status).  IN SHORT, only my system provides a reasonable "containing" theory of behavioral development (and learnings) AND ONLY MY SYSTEM has TESTABLE hypotheses of maturational development and learnings happening at important times (stage shifts), in real effect LITERALLY OCCURRING TOGETHER.  Only my view is biologically grounded as well.  This is why it is a [(THE)] complete and coherent (integrated) outline of learning and development which will serve AI well: it is  internally consistent, has very well definable system of [all]-BEHAVIORALLY-definable elements, AND indicates how the various behavioral elements relate.

Mainstream psychology obviously cannot successful serve AI, OR ITS OWN STUDENTS, for that matter, well.