AI proves that 19x19 is the wrong size for go

General Go Discussion
1

hi, 12k here, i’ve been thinking

if katago and ai always go for the corners, this proves that 19x19 is the wrong size for the game

because in my eyes, ideally, the corners would be just as valuable as the center, and this should be the underlying struggle when playing go

17x17? 21x21?

it would seem the small board, center play is more important?

anyway just a thought
thanks for reading

1 Like
2

Theres been some speculation before that 19x19 provides close-to-perfect balance between ‘3rd line territory’ and ‘4th line influence’

Apparently if you compare the “efficiency of stones”, on 17x17 its considerably more efficient to go for the corners than build center influence, and on 21x21 the center becomes more efficient than making 3rd line territory across the sides

Why 19x19 at Sensei's Library
Classical example of center versus side territory at Sensei's Library

3 Likes
3

thanks for the links, and that’s my point

I wanna see ai play on a 23x23 board

which board size will the computers not value corners so much, ya know?

4

There’s someone who has been experimenting with training KataGo on larger board sizes and to my knowledge they have not found anything different about the best opening moves on larger boards. The corners and the 3rd and 4th lines on the edges are still where the opening takes place, just like on 11x11, 12x12, 13x13, … all the way through 19x19.

The best indications we have with AI point evaluations across board sizes are that the value of playing in the center versus near the edges or corners has little to do with the overall size of the center versus the edges. Thinking that “the center is larger therefore moves in the center should be worth more or be more efficient” seems to be simply the wrong way to think about it.

Here’s a better conceptual picture:

  • Once the board is large enough (anything roughly 11x11 or larger) so that there is an opening at all and not just rapid sharp tactics, then there is an approximate constant baseline value per big opening move owing to the influence that a given stone has on the points around it. This is almost all of the value of that stone, and it doesn’t depend much on board size because it’s mostly a matter of the local value that stone has on the points nearby.
  • A stone that is the ideal distance from edge (3rd or 4th line) gains a small bonus in value due to the edge’s effect on the local tactics, slightly improving the stability of the stone and the control it has over that area around it. This distance also doesn’t depend on the board size because it’s a result of the local tactics and difference in future followups. For example, a 2-point extension on the third line behaves about the same regardless of whether the opposite side of the board is 13 spaces away or 16 spaces away, whereas the third-lineness of it matters a lot for what the future local followups are.
  • Obviously if there are any nearby stones already then the picture changes greatly, we’re just talking about the case when there’s not yet.

This conceptual picture so far predicts/retrodicts our observations much better:

  • The fair komi, and the average value of an opening move, is very similar on every board size starting at 10x10 or 11x11 all the way up through 19x19. (the “center is larger so central stones are more efficient” picture would incorrectly predict the fair komi to vary or scale more, and it doesn’t seem to).
  • The 3rd and 4th lines, and in particular the 3-4 or 4-4 points, are the preferred opening moves on every board size from 11x11 through 19x19 and tentatively for some larger sizes too. (the “center is larger so central stones are more efficient” picture would incorrectly suggest that moves in the center should become more valuable, or that the 5th line might start becoming better on larger sizes, and it doesn’t seem to).
14 Likes
5

because in my eyes, ideally, the corners would be just as valuable as the center, and this should be the underlying struggle when playing go
I think most go players would disagree with that being the criteria for best size.

On 13x13 corners still seem preferable to the center.

You may have to go down as far as 9x9 to make the center move equally optimal as playing (4,4) or (4,3), and we certainly don’t want to limit Go to that size.

Having multiple popular sizes (9x9, 13x13, and 19x19) is one of the great charms of go. You can choose a size according to your playing level, and that of your opponent (beginners should definitely not start with 19x19) and according to how much time you have available and how quick you like to play. Size should not be chosen based on a measure as arbitrary as having the value of starting at tengen be as close as possible to a start at (4,4).

3 Likes
6

I agree, except that I think beginners should try all three sizes if they’re interested. If they want to stick with 9x9, no problem, it’s an interesting game in its own right. If they fall in love with 13x13, have at it, I think that size is very underrated. If they just want to play on 19x19 to get “the full experience”, then more power to them, they can always check out the smaller sizes later

2 Likes