Life In 19x19

In mathematics and computer science there is a kind of algorithm called a "greedy" algorithm. It is an algorithm in which each step consists of choosing the best possible option at the time. There are many examples where the greedy algorithm fails to reach the optimal end result. I wonder whether there could be a greedy algorithm problem with the AI go playing programs. Specifically it might happen that choosing less than optimal moves might result in a "better" win due to synergy between less than "locally" optimal moves.

gowan wrote:In mathematics and computer science there is a kind of algorithm called a "greedy" algorithm. It is an algorithm in which each step consists of choosing the best possible option at the time. There are many examples where the greedy algorithm fails to reach the optimal end result. I wonder whether there could be a greedy algorithm problem with the AI go playing programs. Specifically it might happen that choosing less than optimal moves might result in a "better" win due to synergy between less than "locally" optimal moves.

Using the hill climbing metaphor, you might be climbing a certain hill, but there is another, higher hill that you will never reach, because you would have to go down the hill you are currently climbing. I think perhaps the superiority of the Zero bots, which did not start off copying human play, suggest that humans have been climbing a lesser hill, strategically. The Zero bots found a higher strategic hill to climb. Meanwhile, humans seem to be further up a hill or hills concerned with ladders, semeai, and life and death. Probably also with go infinitesimals, but I can't tell yet.

I think that both the human and AI approaches are capable of finding the highest hill, but doing so could take a long, long time.

Gomoto wrote:Updated 5-3 "Joseki" according to ELF:

(Thanks for your input: drmwc, Bill, Uberdude.)

I stay by my word for now. The 5-3 opening move is very questionable today.
...

It always has been.

ez4u's post reminds me that there is a wealth of information buried away on this forum, if only you can find it. The search functionality does work reasonably well if some long time member recalls a similar thread years ago so can put in some keywords and the author field, but otherwise it's hard to discover. Some other forums have a "similar threads" type of feature which automatically search old posts and suggest them, I wonder if any such phpBB plugins could give useful suggestions (and not overload the server).

Uberdude wrote:ez4u's post reminds me that there is a wealth of information buried away on this forum, if only you can find it. The search functionality does work reasonably well if some long time member recalls a similar thread years ago so can put in some keywords and the author field, but otherwise it's hard to discover. Some other forums have a "similar threads" type of feature which automatically search old posts and suggest them, I wonder if any such phpBB plugins could give useful suggestions (and not overload the server).

With the Participate software back in the '80s we had an admin called the Librarian whose job was to connect similar or related threads. The software was quite flexible, so the Librarian often created new structure. It seems to me that our current structure is complicated enough, but we might have some threads closed to discussion that only provide links. Especially in the Study group. Major topics might be Joseki, Tsumego, Reading, and AI. This thread might be linked from both AI and Joseki, and Joseki might have some internal structure, so that this is linked from 5-3 Joseki.

We need something like a ...

... wiki

senseis library.

Enhance it by AI for automatic placed links

Gomoto wrote:We need something like a ...

... wiki

senseis library.

Enhance it by AI for automatic placed links

Linking to SL is a good idea. And vice versa.

But SL is not so good for discussions, I think. This type of forum is better for that.

On Pandanet IGS many of my opponents recently played 5-3 openings. Tough luck I am prepared

3-3 is still seldom played by pros in response to the 5-3 point.
But here is one example: