Page 6 of 6
Re: Bots that undo
Posted: Wed Feb 04, 2015 10:45 am
by Boidhre
HermanHiddema wrote:Ok, a thought experiment.
Suppose I, a 4 dan player, was paired with a 15 kyu who must, on each move, suggest move candidates to me. Say he suggests 20 candidate moves on each turn, from which I must choose. What do you think our joint playing level would be?
I think it would depend a lot on whether that person was 15 kyu because they consistently badly misjudge the value of moves but have pretty decent for that level ideas for moves or whether they had a decent sense of the value of moves for their level but consistently come up with poor ideas for moves. To take two extremes.
Re: Bots that undo
Posted: Wed Feb 04, 2015 11:01 am
by Uberdude
Boidhre, aren't you somewhere around 15k? Can you suggest 20 black moves here (or did you already look at my suggestions, if so just pick some other random game):
Also if the 15k gets to see which move the 4d chose it's a bit like rengo and that choice can guide the 15k in what to suggest next. My initial answer to the question: I feel could be at least 5k; I would be interested to see results of this experiment
Re: Bots that undo
Posted: Wed Feb 04, 2015 11:07 am
by Boidhre
Uberdude wrote:Boidhre, aren't you somewhere around 15k? Can you suggest 20 black moves here (or did you already look at my suggestions, if so just pick some other random game):
Um, I haven't played much in the last year and a bit. I was around 7k KGS when I stopped, I've no idea of my current strength. Still want me to do it?
Re: Bots that undo
Posted: Wed Feb 04, 2015 11:27 am
by Boidhre
Actually, sounded interesting:
I really dislike some of these moves. "a" would be my choice for what it's worth.
Edit: looking at yours it seems like I missed quite a few obvious moves!
Re: Bots that undo
Posted: Wed Feb 04, 2015 12:08 pm
by skydyr
Uberdude wrote:20 is a lot, I doubt a 15k could do that without resorting to mindlessly naming intersections; hey I would struggle to name 20 plausible moves a lot of the time. The mere act of being forced to think of many moves would likely mean a 15k wasn't 15k by the end of the game. When I think Malkovich players here are being too narrow-minded I suggest they write down just 3 alternative move suggestions. As a test I have picked the most recent Malkovich game here and added my 20:
Now, for our benefit, could you rank order them, or at least group them into classes of largest, next large, medium, small, etc? Ideally with reasons?
Re: Bots that undo
Posted: Wed Feb 04, 2015 2:56 pm
by Mike Novack
quantumf wrote:Mike Novack wrote:
As somebody who designed software in my day (though not game playing software) I can tell you that I would think an AI to find "go reasons" much easier than an AI to try to evaluate "best of the lot" in the context of a particular board position. I can see how fairly simple rules might be able to tell whether this or that go reason applied to a move. Not so for the evaluation process.
If it's that easy, I wonder why MFOGO 12 doesn't do that.?
It
DOES do that. That is precisely what MFOG 12 is doing. It is using an AI "go reason finder" to create a candidate set of moves and then using a MCTS evaluator* to pick the best of those moves. The MCTS evaluator is doing that (evaluating each move) without referring to the reasons attached to that move (by the AI). But after the move has been chosen, the reasons can be displayed. All that requires is having left them attached to the moves.
If you aren't seeing this happen then you haven't turned on that option (to show reasons). Do you have MFOG 12? Or know somebody who does? Playing against a bot over the internet, say on the KGS server isn't going to let you change options.
If you look at the reasons (run with that option on) you can see that a move often has several reasons associated with it. It might be interesting if looking at that list (and of course the move itself in the context of the board position) people did or did not see
other reasons (other "things" the move was doing in the go basics sense). Now that might give us some idea of the "level" of the reasons if we had several people of differing go strength do that. For example:
If looking at the reasons displayed by MFOG 12 a 5 kyu human can't see addiitonal reasons for the move then the reasons are at least the 5 kyu reasons. Similarly for other levels of players. Those of you who think that the reasons shown by MFOG 12 are 15 kyu reasons are predicting that, for example, a 10 kyu player would see additional reasons.
* At its 1 dan and 3 kyu settings. For the 6 kyu setting and weaker MFOG12 is using its old (from MFOG 11) AI evaluator. BTW, Fotland believes he doesn't have the "3kyu" level calibrated correctly, that maybe it is actually 2 kyu. It is really difficult to adjust the MCTS evaluators to be a specific amount weaker than their maximum strength.