Go is well-known to be more difficult than chess for current narrow AI approaches.
This is because most current strategy-game-playing programs work via variants of the "min-max search with alpha-beta pruning" algorithm. This algorithm works by building a tree of future possible game-states and weighting the tree branches as to their likelihood of leading to victory.... Game-specific heuristics are then introduced to weight the leaves of the tree being explored at any given point. Go is not really amenable to this kind of approach. However, this does not imply that Go requires an AGI. It may be that a different sort of narrow AI approach could succeed in creating a world champion computer Go player, without significant general intelligence. In fact, that is my intuition. -- Ben G > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On > Behalf Of Pablo Carbonell > Sent: Wednesday, January 22, 2003 8:03 PM > To: [EMAIL PROTECTED] > Subject: Re: Re[2]: [agi] Turing Tournament > > > Hi, > > Anyone knows about the chinese game called "go"? > > I read once in a magazine that a rich man in Asia is willing to > pay 1 million dollars to the one who builds a program that > wins against the best GO players. > > What I really like from that game is that it looks very simple > (the rules are extremely simple!) but thinking of a move gets > very complex. > > Maybe it's already done, I don't know... > > What do you think, Alan? > > Cheers, > Pablo > > > > > Wednesday, January 22, 2003, 9:33:16 PM, Kevin Copple > wrote: > > > > KC> I spent a few minutes looking at the CalTech Turing > Tournament website > > KC> http://turing.ssel.caltech.edu/index.html I came away > rather puzzled. This > > KC> seems to be a number guessing game. Sure, it > includes both emulator and > > KC> detector algorithms, but such a specialized domain > seems less interesting > > KC> than algorithms that play chess, bridge, go, or > whatever. > > > > I was pretty puzzled at the game they proposed as well. > > > > However, the games you recommend probably are too > large and complex to > > build such a tournament around, especially when including > detectors. > > > > Without actually sitting down and playing through the > proposed game a > > number of times, it's difficult for me to see how exactly it > would > > work...maybe the point is to detect how humans learn the > game, but in > > that case a playing program could be written to slowly > converge on a > > good strategy. But with those kind of "meta-strategies" > involved, the > > small details (how much knowledge will the human players > have of the > > game? do they get to practice? how are they selected -- > random > > students? etc.) get to be critically important, and the end > result > > seems like it would be a crapshoot. > > > > In principle the idea of building imitators and detectors and > setting > > them against each other *sounds* neat, but when you get > down to > > specifics things become muddled, at least for me. It might > make more > > sense as two separate tournaments. > > > > -- > > Cliff > > > > ------- > > To unsubscribe, change your address, or temporarily > deactivate your subscription, > > please go to http://v2.listbox.com/member/? > [EMAIL PROTECTED] > > > > > ----------------------------------------------------------------------- > Registre su dominio. El primer paso de un proyecto exitoso en Internet. > http://www.montevideo.net.uy/hnnoticiaj1.exe?9,0 > > ------- > 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]
