Thursday, July 12, 2007, Cenny Wenner wrote: CW> There are certainly difficulties with natural language but I do not CW> see how these empirical and practical difficulties can be called CW> tautologies of languages in general.
Problem here is not in certain deficiency of natural language or some mystical way goals always keep being misinterpreted. Problem is that if you want to formulate complex goal/value, it usually intrinsically employs informal everyday concepts, and these concepts exist within the system you want to instruct only as results of complex process of perception. If you want guarantees on overall result, you must include properties of this perceptual process in specification as well. It draws practical line between goals we can build systems to reliably achieve and those we can't. There's obviously a technical problem of building a system powerful enough to be able to recognize these concepts, but it doesn't help in verification of whether these cancepts are really recognized in those situations we want them to or not. Problem is twofold: you should be able to guarantee properties of this perceptual process (so it can't be an arbitrary emergent one), and you should be able to figure out the essense of your own perception of these concepts. Middle ground is to make simplier formulation of goals not involving too much hairy perception. Meta ground is to include human perception itself in the loop, thus tasking the system to figure out perceptual details of humans as a subtask. -- Vladimir Nesov mailto:[EMAIL PROTECTED] ----- 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=4007604&id_secret=20587889-3be28f
