#Unbeatable chess ai series#
After attendingīruce''s excellent lecture, I got into a usenet discussion with someone who thought thatĭesigning and building AI for computer wargames was easy (of course he had never done one)Īnd during the conversation I offered up the above derivitive to Bruce''s orginal title.Supporting the recent launch of AMD Ryzen™ 7000 Series processors, the new AMD AM5 Chipset offers users high power performance across gaming, creating, and intensive computing tasks. "Chess is Easy, Go is Hard", in which he discussed his AI work for the game Go. While working on the game ArmyMen presented a lecture at GDC entitled In the interest of giving credit where credit is due, Bruce Wilcox (at 3DO at the time), To paraphrase Eric from a GDC several years back, "Chess is hard, Go is harder, wargames are hard cubed ".
I would argue that the most simple wargame, however, is infinitely more complex than either. The terrain is flat and the complexity comes from the way the pieces move and the constraints of the board. J.R.R TOLKIEN ACHIEVES 40,000 RPM IN GRAVEįor the record, games like Go and chess are extremely complex but they''re clearly "solvable". You should be able to do with a bit less then 1 terrabyte of memory ^_^ When the enemy is in the lead, they try to "dismember" the enemies formation, buying them enough time to get to their location. When they are "in the lead", they try to attain their positions. When the enemy takes a move, they just calculate the amount of time it would take the enemy to get a favourable position (more or less), and compare it to their own. Usually, they aren''t fazed much by what the enemy does.
Every player has his own set of preferred formations, but most good chess players know a lot of them. You try to move your own pieces into such a formation that the enemy can''t do a thing. Thats about the same approach as a real chess player uses (or at least, those that I know).
When the player moves a "different" piece, the computer would just have to search for a new "favourable" set that can be attained, and move towards that set. And then you wouldn''t have to store them all, but rather store a "few favourable" positions that the computer should try to attain at certain points during the game. And really, you can cut down a *lot* of ''possible'' sets, becuase they are either impossible to get in-game, or terribly unlikely (even because it would put the computer at a disadvantage). Eventually ^_^ Thing is, you shouldn''t look at the "game" that is played, or the moves that have been made, but rather towards possible layouts of the board. To paraphrase Eric from a GDC several years back, "Chess is hard, Go is harder, wargames are hard the High, Cold, Snowy Mountains of Colorado Go isn''t there yet but it will follow without a doubt.įor the record, games like Go and chess are extremely complex but they''re clearly "solvable". That doesn''t mean that an AI can''t be written to beat 99.99% of players-look at any "average" chess program. There are a thousand ways to approach the game but no way (yet) to truly "master" it. I put chess in the "solved" category with checkers and tic-tac-toe more because there have been so many papers and treatises written on the subject.
Tricron2 might have gotten the "solved" designation from me on my site. As for the statement it wasn''t a joke, and I think i found it on gamedev.