English Chess Forum

Ten years ago, AlphaGo announced its arrival by winning a match against one of the world's leading Go players. This week the Korean press has been claiming that Humans Strike Back when another Korean won a match against "the world's top" Go AI programme. A more detailed account revealed that not only was the human (Shin Jin-seo) given a two stone handicap, but a seemingly massive time advantage. Does someone who understands Go (which I don't) know whether this result is in any way significant?

I don't know much about Go either. When you say he was given a handicap, does that mean an advantage (as in a golf handicap) or a disadvantage (as in a horse racing handicap) ?

An advantage: if I've got it right, he could place two stones on the board before the game started. The implication in the articles is that this is a significant but not overwhelming advantage, maybe the equivalent of pawn odds rather than rook odds, at a guess. The human was also given much more time on his clock. In the original match ten years ago, they played as equals - as with Kasparov against Deep Blue - and AlphaGo won 4-1. Go programmes are supposed to be much stronger now than then, so it seems this result was a surprise.

See https://www.britgo.org/about/rating

which seems to be what Pete says. I know a (human) Go player and will ask him if he can explain in such a way that I understand...

I also know a human Go player and he explained this to me in a way that was really helpful (the difference between "stones" and "points" advantages). The essence is that a top human Go player should be able to beat a strong program when the human has this two stone advantage (and this result seems to confirm that). In other words, this is a way of measuring the (currently unknown) gap between top human players and top Go programs.

He did make the interesting point that when a pro has an advantage like this that is slight (but not overwhelming), they might choose to gradually give the advantage back for the sake of simplifying the game. Reminds me a bit of those chess positions where strong players will give back material to simplify into a winning endgame.

Translate two stone go handicap into chess terms.

AI Overview "A two-stone Go handicap translates in Chess to pawn and two moves odds or a rating difference of roughly 200–300 Elo points."

AI goes on to tell us "Pawn and two moves: The weaker player plays Black, removes White's f7-pawn (or b7-pawn depending on tradition), and makes two successive moves before White begins" so it may not know quite as much about chess as one would hope.

David Williams wrote: ↑

Thu Jul 30, 2026 8:28 pm

AI Overview "A two-stone Go handicap translates in Chess to pawn and two moves odds or a rating difference of roughly 200–300 Elo points."

AI goes on to tell us "Pawn and two moves: The weaker player plays Black, removes White's f7-pawn (or b7-pawn depending on tradition), and makes two successive moves before White begins" so it may not know quite as much about chess as one would hope.

It's known that AI makes it up as it goes therefore not to be trusted on factual matters.

A 200 Elo point difference translates to about a 3 in 4 advantage to the higher rated player and a 300 point to about 7 in 8. Is this right for a two stone handicap between players of otherwise equal strength?

AI - the subtle enshittification of humanity's ability to reason.

We've now come full circle, as my original post was about AI failing to defeat a human in a short Go match - when it had to give the human (Shin Jin-seo) a significant starting advantage. Despite this loss, it seems that AI is better at Go than at giving help to googlers.

"Despite this loss, it seems that AI is better at Go than at giving help to googlers."

It's almost as if AI is trying to please googlers, by giving them an answer, even if it's made it up. That matters, but if it loses a game of Go, it doesn't matter particularly, as it's impressive that it can play such a game to a high standard.

I hope AI doesn't object to the appellation "it".