# AlphaGo and AI **URL:** \ **Category:** Factual Questions\ **Created:** [July 4, 2025, 1:21am UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466 "2025-07-04T01:21:31Z")\ **Posts on this page:** 13\ **Page:** 1
**Author:** ![MikeF](https://avatars.discourse-cdn.com/v4/letter/m/45deac/32.png) [@MikeF](https://boards.straightdope.com/u/MikeF)\ **Post date:** [July 4, 2025, 1:21am UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/1 "2025-07-04T01:21:31Z")
I just gone done reading a book, _The MANIAC_ by Benjamin Labatut. Its a fictionalized account of the life of polymath Jonn von Neumann. Much of it is based on fact. It ends with the true story (at least, the outcome) of a computer, AlphaGo, that beats the worlds best Go player in 2016.Each follow up computer is better than the last with the new one beating the old one every time. .Most of the science is way over my head but in the end, the book is about AI and where it is heading as well as the world of some of the most brilliant people My question is this - If two identical computers like this played each other, what would happen? A draw every time? 50/50 record for each? BTW, I really liked and highly recommend the book. ---
**Author:** ![Whack-a-Mole](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/whack-a-mole/32/141_2.png) [@Whack-a-Mole](https://boards.straightdope.com/u/Whack-a-Mole)\ **Post date:** [July 4, 2025, 1:56am UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/2 "2025-07-04T01:56:10Z")
In fact, AlphaGo has played itself quite a lot. That is how they trained the AI model. First they showed AlphaGo records of human games to learn from but then had it play itself to train it. The results are not a tie because AlphaGo is not playing off a list. In other words, it is not seeing board configuration A and deciding the only move must be B. Instead, it is trying to figure out the next best move (or series of moves). That is how it improves a little each game it plays. The two AI will not get the same answer every time and, as they diverge, each game becomes unique. ---
**Author:** ![Schnitte](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/schnitte/32/9033_2.png) [@Schnitte](https://boards.straightdope.com/u/Schnitte)\ **Post date:** [July 4, 2025, 12:15pm UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/3 "2025-07-04T12:15:47Z")
> [@Whack-a-Mole](#): > > The results are not a tie Go games, at least under the rules that usually apply in high-level games, cannot end in a tie (except for very rare circumstances, such as a game that has to be annulled). In a go game, players score points by securing territory and capturing opposing players; but there is almost invariably a handicap (called a _komi_) which is added to the score of the player playing second (white), to compensate for black’s advantage of the first move. This komi is usually 6.5 or 7.5 points, but since you can only collect an integer number of points during gameplay, a tied score is not possible. Something very similar to what the OP is having in mind is happening in top-level chess, where the majority of games end in a tie nowadays. The final of the 2024 world championship between Gukesh Dommaraju and Ding Liren, for instance, consisted of 14 games, nine of which ended in tie. Gukesh won three of the remaining five and Ding two. At that level, and with all the theoretical study that has been happening in chess, the differences between players are so slight that a tie becomes sort of the default outcome of a match. ---
**Author:** ![Snarky\_Kong](https://avatars.discourse-cdn.com/v4/letter/s/a183cd/32.png) [@Snarky\_Kong](https://boards.straightdope.com/u/Snarky_Kong)\ **Post date:** [July 4, 2025, 6:28pm UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/4 "2025-07-04T18:28:32Z")
In computer chess competitions, played from the standard starting position, I believe every game ends in a tie. So why are there competitions? Well, they don’t start in the standard position, play the Berlin, then repeat moves. They start several moves in, okay the game then swap sides and play the same start position. But in general, do equally matches machines have 50% win rates against each other? I mean, how else would you call them equally matched? ---
**Author:** ![Schnitte](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/schnitte/32/9033_2.png) [@Schnitte](https://boards.straightdope.com/u/Schnitte)\ **Post date:** [July 4, 2025, 6:42pm UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/5 "2025-07-04T18:42:54Z")
> [@Snarky\_Kong](#): > > But in general, do equally matches machines have 50% win rates against each other? I mean, how else would you call them equally matched? In all sequential move games, there’ll be an inherent asymmetry between the players in the sense that it’s either an advantage or a disadvantage (depending on the game - in chess and go it’s an advantage) to play the first move. Go tries to compensate for that by means of komi (although there is still a debate as to the correct quantification of that komi), whereas in chess this is not done. If two equally strong machines are matched against each other, you’d expect white to win. Not every single game, but in the long run there’d be a statistically significant lead for white. ---
**Author:** ![Snarky\_Kong](https://avatars.discourse-cdn.com/v4/letter/s/a183cd/32.png) [@Snarky\_Kong](https://boards.straightdope.com/u/Snarky_Kong)\ **Post date:** [July 4, 2025, 6:44pm UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/6 "2025-07-04T18:44:20Z")
I don’t think it’s proven that white can force a win. And in any case, that’s why competitions switch sides and equally matches competitors will have 50% win rates, which is exactly what I said. ---
**Author:** ![Hari\_Seldon](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/hari_seldon/32/5173_2.png) [@Hari\_Seldon](https://boards.straightdope.com/u/Hari_Seldon)\ **Post date:** [July 4, 2025, 7:50pm UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/7 "2025-07-04T19:50:16Z")
Really? When I was playing the standard Kome was 4.5 points. They told me it should be 5.5 but that would lead to dull games and this way white would have an incentive to complicate the game. In any case, draws are not possible because of that 1/2 point. ---
**Author:** ![Schnitte](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/schnitte/32/9033_2.png) [@Schnitte](https://boards.straightdope.com/u/Schnitte)\ **Post date:** [July 4, 2025, 9:05pm UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/8 "2025-07-04T21:05:06Z")
There’s a dispute as to what the correct komi should be, but the prevailing view today is that it would be somewhere around 6 or 7; in the past, lower figures would have been preferred. Sensei’s Library has an extensive discussion of the topic: [Komi at Sensei's Library](https://senseis.xmp.net/?Komi) ---
**Author:** ![Whack-a-Mole](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/whack-a-mole/32/141_2.png) [@Whack-a-Mole](https://boards.straightdope.com/u/Whack-a-Mole)\ **Post date:** [July 5, 2025, 11:36am UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/9 "2025-07-05T11:36:40Z")
Is chess a “solved” game that a computer can look at any possible board configuration and, no need to think about it, knows the next best move? All solutions are basically on a very big searchable list? ETA: I am reasonably certain that the solution set for Go is waaaaaay bigger than chess. Stupidly big (and chess is also big, just not that big). ---
**Author:** ![pulykamell](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/pulykamell/32/3166_2.png) [@pulykamell](https://boards.straightdope.com/u/pulykamell)\ **Post date:** [July 5, 2025, 11:58am UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/10 "2025-07-05T11:58:08Z")
No. Chess is solved up to, I believe, 7-piece end games (including kings). ---
**Author:** ![Whack-a-Mole](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/whack-a-mole/32/141_2.png) [@Whack-a-Mole](https://boards.straightdope.com/u/Whack-a-Mole)\ **Post date:** [July 5, 2025, 12:02pm UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/11 "2025-07-05T12:02:40Z")
You’d think the solution set would get smaller and more easy to compute as it gets to the endgame. I admit that is a gut reaction. I really do not know. Since it has not been done (yet) clearly there is complexity I am missing. (Put another way, what am I missing?) ---
**Author:** ![pulykamell](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/pulykamell/32/3166_2.png) [@pulykamell](https://boards.straightdope.com/u/pulykamell)\ **Post date:** [July 5, 2025, 12:06pm UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/12 "2025-07-05T12:06:33Z")
I don’t know what you mean by this. The solution set does get smaller the fewer pieces you have left on the board. 2 pieces is obviously tie (2 kings). 3 pieces should be easy to compute wins and ties for. 4 pieces you have more complexity. 5 even more. 6 even more. 7 is as far as we have gotten when it comes to endgame lookup tables. [Here’s something I found with how big these tablebases are](https://en.wikipedia.org/wiki/Endgame_tablebase). For an 8-piece, there are an estimated 38 quadrillion positions. For seven, 423 trillion. For six, 3.7 trillion, 5 or fewer is 26 billion, combined. ---
**Author:** ![Whack-a-Mole](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/whack-a-mole/32/141_2.png) [@Whack-a-Mole](https://boards.straightdope.com/u/Whack-a-Mole)\ **Post date:** [July 5, 2025, 12:17pm UTC](https://boards.straightdope.com/t/alphago-and-ai/1020466/13 "2025-07-05T12:17:22Z")
> [@pulykamell](#): > > I don’t know what you mean by this. I think I misunderstood what you were saying. I was thinking the board was counting down to seven. Not that we solved backwards (i.e can we solve for 2, 3, 4 left on the board as opposed to starting at the beginning and can we solve for 32, 31, 30…). I should have given it more thought. My fault. Thanks for the explanation.