5 ms·
Hi, I'm a Go player (only around 9 kyu though). The programs don't make "intuitive" moves, but nor do they exactly perform brute search - the search tree is fa
by Wildgoose 11y ago
Hi, I'm a Go player (only around 9 kyu though).
The programs don't make "intuitive" moves, but nor do they exactly perform brute search - the search tree is far too wide.
They actually use a mixture of techniques. For example, in the beginning they may select moves from fuseki databases of accepted good play, and have some knowledge of joseki for plays in the corner.
During the very end game they may be able to brute force search and make sensible decisions about the order of play - however there is also a book about using simulated annealing with neural networks to decide the order moves should be played. <a href="http://www.amazon.co.uk/Mathematical-Go-Chilling-Gets-Point/dp/1568810326">Mathematical http://www.amazon.co.uk/Mathematical-Go-Chilling-Gets-Point/... Go: Chilling Gets the Last Point</a>.
There have been attempts to code "rules of thumb" into the Go algorithms, such as knowledge of bamboo joints, table shapes, etc., but the recent improvements have largely been down to techniques such as evaluating the position after allowing two consecutive moves and even simply trying various random plays rather than an exhaustive breadth-first search.
There certainly used to be a mailing list for people interested in the development of Go-playing programs. I can't remember the address but if you broaden your search terms to include "Go, wei-chi, weiqi, baduk" you should be able to find it.
- gjm11 11y ago> using simulated annealing with neural networks Are you sure about that, or are you basing it on the word "chilling" in the title? My understanding is that Berlekamp's book is all about applying combinatorial game theory to go, and the "chilling" in question has very little to do with the "temperature-decrease" schedule in simulated annealing. "Chilling" in combinatorial game theory means, in effect, making players pay a penalty for each move they make. The simplest games to analyse are ones in which it's always a disadvantage to have to move; these are equivalent in a certain sense to numbers; the "chilling" operation tends to turn games into simpler games. See e.g. http://senseis.xmp.net/?Chilling http://senseis.xmp.net/?Chilling for a little more information. The only connection between this and what happens in simulated annealing is that temperature happens to be a good metaphor for both. In combinatorial games, that's because the "value" of a position changes less wildly with each move in a cooled/chilled game. In simulated annealing, it's because the configuration you're looking at changes less wildly on each successful update as the temperature decreases.
- CaiGengYang 11y agoSounds very interesting ! God knows how Go AI will evolve in the future ! Perhaps the field of Go AI will produce a fully conscious machine in the future , one that can think and strategise like a human being , contributing to the technological singularity predicted by Ray Kurzweil ...
- Wildgoose 11y agoSorry, yes, you are correct - my memory was playing tricks.
- CaiGengYang 11y agoPlaying tricks about what ?