Regarding "how to select the appropriate reasoning rules to apply" --- in
Novamente this occurs on two levels:
1)
some simple heuristics applied as a default
2)
based on probabilistic rules that are learned based on experience (via the
system's experience carrying out reasoning)
Note
that there is some recursiveness here, because these probabilistic rules
regarding which inference rules to apply in which context, may also be learned
via INFERENCE. But this leads potentially to an infinite-depth recursion
of inference guided by rules learned by inference guided by rules learned by
inference ... which is why one needs some simple heuristics at the bottom to end
the recursion after a finite number of steps.
-- Ben
G
-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]On Behalf Of Yan King Yin
Sent: Thursday, September 08, 2005 8:06 AM
To: [email protected]
Subject: Re: [agi] Representing ThoughtsI didn't show how the reasoning itself would be done in Novamente because my time was limited and the trains of reasoning would be pretty long!We haven't yet tried NM on this kind of example but plan to do so in early 2006. This fall our main AGI goal is to get NM to automatically learn Piagetan "object permanence" based on experience. Conservation rules like conservation number come after that on our Piagetan learning trajectory.However, we have worked out some inference trains of comparable complexity. But to show you these inference trains would be tricky because it would involve a lot of highly technical NM-inference-module terminology and notation.Regarding how the reasoning is done: Yes, in Novamente the reasoning is spontaneous. Regarding the general problem of how to get logic systems to do spontaneous inference, the real problem is defining what is a "good conclusion." It's easy to get a logic system to produce consequences of its premises, but, the question is how to get it to direct its course of inference toward the "interesting" conclusions. One needs an "interestingness" metric, then one can use an inference-control heuristic such as tapered N-best search in a way that uses the interestingness metric to assess "best"-ness. NM's current interesting metric values predicates with surprising probability (based on the system's prior knowledge), high amount-of-evidence, and small size (compactness). So, when acting in spontaneous non-goal-oriented mode, Novamente "spontaneously" takes its premises and tries to carry out inferences that will lead from its premises to conclusions satisfying these merits.re Spontaneity:
Sorry that I raised the issue, but I think spontaneity is actually nonessential to AGI. The most urgent need is to build an AGI that can answer queries or achieve goals. Autonomous thinking is nice to have, but not essential.If the thinking is goal-directed, then the problem is how to select the appropriate rules to apply (such as the rule of numerical conservation) among possibly thousands or millions of rules. What we need is perhaps an associative mechanism.re Numerical conservation and Piagetian learning:The more important issue is how to learn Piagetian abilities from experience. I think the key to this kind of learning is "inductive" in a broad sense (not the same as mathematical induction). For example, if the AGI encounters several blondes who are dumb, it may draw the conclusion that all blondes are dumb. My theory is that a lot of basic rules of abstract reasoning emerge in the brain this way, but we are not conscious of it. A small child may understand that if there are 2 apples and if he has 2 elder brothers, then he may be left with no apple. It is from numerous experience of this kind that the rule of numerical conservation gets established in the brain.
I cannot think of any way other than induction, through which the numerical conservation rule or other Piagetian abilities can emerge. But if you ask a philosopher like Kant, he may think the rules of logic are of a different category from the rules generated from experience (eg blondes are dumb). As far as practical matters go, however, I think all rules of thinking are derived from experience.yky
To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
