Of course, if you collect valid statistics

Just to further articulate: A basket of apples costs 10 euros. The same basket of oranges costs 20 euros. The same basket of pineapples costs 30 euros. How much does the same basket of pears cost?

We need to define what means the "solve"? We mention strong AI or weak one? It is necessary to answer for one question, a pool of similar questions to solve?

Dr. Gajniyarov,

First of all, thank you for getting involved in the discussion of this fascinating question. I see your points very well in both of your responses. And I also see that somewhere we have a resonance, but let us further elaborate on it.

Having said that, I would like to emphasize that not the type of problem to be solved is interesting for me, but the view of the people on the opportunity of "solving" (whatever it means) not fully deterministic (or non-deterministic) problems (whatever it means). At the time of posing the question I though of systelligence (partly implanted, partly acquired system intelligence) that makes smart-cyber-physical systems capable to achieve their working/operational/servicing objectives. Systeligence is a different genre for me, than AGI or ASI. Where we resonate is about what is hiding behind your fist answer, but also partly behind the second one. Namely, current effort of AI research (as well as of cognitive engineering) is to reproduce some pattern manipulation capability of the human brain (or better to say, the human mind). The argumentation behind is simple - cognitive or whatever patterns can almost always be found. This applies independent of the fact if a pattern is strong or weak (like our mental models), or if it is concrete (as the mind images generated by sensing) or implicate (as generated by abstraction or illusion). It is also not making a big difference if the pattern is intended or not. Cognitive scientists have proven that our brain/mind tends to look for patterns in what we encounter. It uses some preferred patterns to recognize, describe, explain, predict or control some related phenomenon, as well as to create associations. This is a kind of innate of how we think and the current trend, as I noted above, is to carry it over to artificial constructs.

Coming back to my initial question, the issue is that how much we can reconstruct it and, more importantly, will AI be able to step over what has been identified as limitation for human being. I am doubtful, but open for any discourse. Why I am doubtful, or even sceptic? Easy to explain. In the simplest form it is about how to resolve the so-called multi-pattern-match situation. In both of the examples I mentioned this is what we are facing. There is no one concrete (ultimate) solution, and to get to any meaningful solution we need strong abstraction. In these and similar quizzes, deduction, induction, retrospection, or resembling other formal reasoning strategies (including probability or statistical significance) will not work. Obviously, a computer system can use abduction and may find the most probable answers, it does however not mean that it will be the absolute one. Furthermore, if the need for multi-step abstraction is present, then I guess we really have to reconsider everything. Some tasks or questions, like this presented ones, not only imply multiple, alternative, (undecidedly) equal solutions and answers, respectively, but also need abstraction (out-of-the-box thinking). For the computer the issue is: (i) to find the relevant patters by abstraction, (ii) to select the most relevant patterns in the context, (iii) rank the most patterns based on criteria other than just relevance (other semantics, constraints, intuition), and (iv) if it fails, do not hesitate to refer to Gödel's theory of undecidable.

So, again, coming back to the original question: If there are undecidable logical (or whatever) problems for humans, what is the basis to assume that these problems will not be out there for (any kind of) artificial intelligence-enabled artificial cognitive systems as intractable?

Kind regards,

I.H.

Dr, Horvath,

Thank you for such a detailed answer. Question matter become more clearer.

In this case I follow a idea expressed by Henri Poincare in the book "About science", ISBN 5-02-014328-6 (rus translate, sorry that i don't have detail reference for en book): machines are powerful, but for truely they lack a spatial degree of freedom and self-replication. Thus, this is a necessary condition for AI to solve such problems.

It is very pleasantly to discuss this issue with you.