AI Opinion on Handicap Placement?

Has anyone ever tried getting AI to evaluate the best placement of handicap stones in a free-placement scenario?

My first thought was to investigate this just by progressively passing as white and seeing what the AI recommends for black. But this isn’t quite right. For example, it might be that the best way to place 4 stones is on the 4-4 points. But if you know that you’ve actually got 5 stones, maybe you’d place the 4th move on a 3-4 point so that you could get a stronger enclosure with the 5th.

Anybody know of any AI investigation of this topic?


I share that opinion that a simultaneous play of the X stones could give a different result as one after the other. Now i never heard about research on this ideal placement although it can be really interesting.


We had a thread about this, let me search…


Well, it seems we don’t really have a thread about this, but there’s some investigation to be found in this topic (post 23 and after):


From that post in the thread, AI and placement start to be the suject of the discussion.

(Thx @Vsotvep for the link)

To resume

In term of points difference between free and fixed, that’s nothing so important but some free can be much worse
In term of usefulness, more should be considered as only the points value, like solidity, easyness to handle, fitting the play style, etc…


Thanks both for the link!

Seems kind of surprising that there’d be nothing better than the traditional placement given AI’s typical preference for enclosures over side star points. But I guess this is just hard to detect since the win percentage so quickly approaches 100% (hence the appeals to score estimator in that thread).

I wonder if one could get more meaningful results by actually pitting weaker AI against stronger ones in handicap games.


You are assuming that “better” is something that benefits the player receiving the extra stones, rather than a placement that would promote an even game. In my observation, extra stones beyond an ambiguous breakpoint are much more valuable than the early stones. In other words, stones 7 to 9, for example, usually have a lot more value individually than stones 1 to 3. It is hard to say exactly where this effect takes over, probably because it depends on the general level of the players. It is conceivable that the traditional placement arose because this effect was instinctively recognized, leading to a deliberately less-than-optimal placement of the later stones.

There is something not taken in account by AI and curiously not explicitly mentioned in both threads about the traditional placement vs a free placement.

That is the sum of specific knowledge on traditional handicap vs the quasi no existence of writing about free placement.

Free placenent is like being an explorer coming with your teacher in a new land incognita. At reverse, traditional placement is like taking holidays in the swiss mountains with a bunch of well maintained hiking tracks.

What are the consequences in the matter of best placement? If it’s about chance of winning free placement could be better because the stronger doesn’t have that settled catalog of strategy/tactics to use. But if it’s about teaching, the traditional should be better as it helps a lot the stronger who knows already a bunch of things to transmit and can more easely answer the student.

Of course and it has already been mentioned, with free placement you still have to find a coherent way in chosing their placement and the next steps to use them well.

In any handicap game it’s crucial to take care from the very beginning of these handicap stones because if not they will just compensate for the many failures you’ll make, like sinking in the flow of the game, and that won’t be enough

1 Like