Yes there's no reason why the fitness criteria or "rules of the game" can't also be subject to exploration during play. I suppose this is "dangerous" only in the sense that an AGI's high level or top level motivations could be subject to change during playful characterization of the space of possible actions.
If my understanding from viewing last year's AGIRI workshops is correct an AGI's top level motivators are supposed to be sacrosanct and immutable, in order to prevent the system from going "unfriendly". I don't think free will is really a problem. During play a system may discover more than one attractor basins for a particular type of task or a particular rules/task combination (often a problem can be solved in more than one way). The decision to select a particular attractor then becomes a far more sensitive one, subject to initial conditions. Flipping between attractors is still a deterministic process, but is at least exposed to the possibility of being sensitive to initial conditions. On 06/07/07, Mike Tintner <[EMAIL PROTECTED]> wrote:
Bob M: >I think the purpose of play is that it allows the system to search the > space of possible actions in a broad yet shallow way, and characterize > the landscape under various fitness criteria. Careful. What you, Bob, "a" & Vlad are all. ahem, playing with here, innocently perhaps, is the worst kind of heresy. For one thing, play is free not determined. As I insisted a while back, to intense opposition from the doyens of AGI, we humans (along with animals) are indeed free, and designed to be free precisely in order to explore (or play). And an AGI must also be free if it is to succeed and survive - .by exploring unfamiliar environments.. Worse still, if we or any AGI can truly play, then we can't actually be programmed in any conventional, only the most minimal sense. For example, Bob, if you carried your logic through, you would realise that an agent that can play, will not only be able to play with its immediate actions within an activity, but also with its "fitness criteria" - its values and rules, just as humans are actually able to do. Children, for example, play around quite explicitly and consciously with the rules of their games, (as do adults). There is a fundamental opposition between play and the conventional program, both in narrow AI and current AGI. So this is dangerous stuff - it challenges the foundations of cog sci & AI - that we and any AGI are, and must be, 1) deterministically 2) programmed.. You could call this heretical tendency not so much the idea that we are a "blank slate" pace Pinker, as a "blank page." Keep on playing. ----- 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/?&;
----- 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&id_secret=13154093-b9a44d
