Mike, I believe many of the confusions on this topic is caused by the following "self-evident" belief: "A system is fundamentally either deterministic or non-deterministic. The human mind, with free will, is fundamentally non-deterministic; a conventional computer, being Turing Machine, is fundamentally deterministic". Based on such a belief, many people think AGI can only be realized by something that is "non-deterministic by nature", whatever that means.
This belief, though works fine in some other context, is an oversimplification in the AI/CogSci context. Here, as I said before, whether a system is deterministic may not be taken as an intrinsic nature of the system, but as depending on the description about it. For example, NARS is indeed "nondeterministic" in the usual sense, that is, after the system has obtained a complicated experience, it will be practically impossible for either an observer or the system itself to accurately predict how the system will handle a user-provided task. On the other level of description, NARS is still a deterministic Turing Machine, in the sense that its state change is fully determined by its initial state and its experience, step by step. Now the important point is: when we say that the mind is "nondeterministic", in what sense are we using the term? I believe it is like "it will be practically impossible for either an observer or the mind itself to accurately predict how the system will handle a problem", rather than ""it will be theoretically impossible for an observer to accurately predict how the system will handle a problem, even if the observer has full information about the system's initial state, processing mechanism, and detailed experience, as well as has unlimited information processing power". Therefore, for all practical considerations, including the ones you mentioned, NARS is nondeterministic, since it doesn't process input tasks according to a task-specific algorithm. [If the above description still sounds confusing or contradictionary, you'll have to read my relevant publications. I don't have the intelligence to explain everything by email.] Pei On 5/6/07, Mike Tintner <[EMAIL PROTECTED]> wrote:
Pei, Thanks for stating your position (which I simply didn't know about before - NARS just looked at a glance as if it MIGHT be nondeterministic). Basically, and very briefly, my position is that any AGI that is to deal with problematic decisions, where there is no right answer, will have to be freely, nondeterministically programmed to proceed on a trial and error basis - and that is just how human beings are programmed. (Nondeterministically programmed should not be simply equated with current kinds of programming - there are an infinity of possible ways of programming deterministically, ditto for nondeterministically).
----- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=231415&user_secret=fabd7936