Playing against bots?

[MCSens]

Hello everyone

What do you think of playing against bots as a beginner?

Thanks in advance!

[illluck]

I personally found it's more interesting to play humans, but I can't imagine how playing bots would be detrimental. Just make sure the program you are choosing is not too weak.

[Uberdude]

Bots aren't very good at friendly banter or reviewing the game afterwards with you.

[Inkwolf]

As long as you play humans as well, practice against bots can't hurt too much.

Once you get stronger, and are playing a bot of near to equal strength, you will discover specific weaknesses in bot games, and be tempted to take advantage of them...even though a human player would have been likely to play differently. For example, I have probably had a hundred wins against Gnu Go 3.7 that ended like this. Chances are, a human would have tried harder to make the trapped stones live, and may well have succeeded. (My human nemesis always manages to find life when I try this pattern against him.) After a lot of wins in this general pattern, it only turned into a contest to see how fast I could make the computer resign (48 moves is my best!) (NOTE: I know this is TERRIBLE Go, and nobody has to tell me that. Any respectable human would never play like this.)

To break the bad habit and get some benefit from bots again, I now play different game engines and now only play Gnu Go 3.7 as white, which is quite a different experience.

One other bad habit to be wary of from bot playing is the habit of clicking quickly without reading ahead. The computer goes fast, leading you to do so, as well, and since there are no consequences for wanting an 'undo' you can get into sloppy habits.

http://www.lifein19x19.com/forum/download/file.php?id=2777

[Samura]

In the beginning, I was shy of play with humans, afraid of being a too weak player. So, I started in Go playing against Igowin HD (Many Faces of Go for iPad) hundreds of games. My experience is that you learn to avoid a lot of basic mistakes, mostly about patterns in fights in the first 3 lines of the board (probably every beginner understand what I mean).

But that is it. When I started to play at KGS, I remember being destroyed by 21kyus while beating consistently Igowin HD set as 15k. The moral of the story is: bots are very good to acquaint itself with the basics of the game, but not much more than this.

My theory as that bot players (at least those that use Monte Carlo methods) don't overplay, ever, and thus they have a very passive style. But as a beginner, you will play against other beginners, which means you will be straight at "Overplay Town". If you don't learn how to play against overplays and aggressive players, you'll never pass the beginner level.

[PeterPeter]

My theory as that bot players (at least those that use Monte Carlo methods) don't overplay, ever, and thus they have a very passive style.

The Aya engine used in this app (https://play.google.com/store/apps/deta ... 3J5LmdvIl0.) often tries unwise invasions, at the lower levels at least. Some of the user reviews are critical of this. I think it is good practice for knowing when an invasion needs a response, and when it can be ignored. Under Japanese rules, you have to decide this for yourself, rather than just throwing stones at everything.

[Mike Novack]

In the beginning, I was shy of play with humans, afraid of being a too weak player. So, I started in Go playing against Igowin HD (Many Faces of Go for iPad) hundreds of games. My experience is that you learn to avoid a lot of basic mistakes, mostly about patterns in fights in the first 3 lines of the board (probably every beginner understand what I mean).

But that is it. When I started to play at KGS, I remember being destroyed by 21kyus while beating consistently Igowin HD set as 15k. The moral of the story is: bots are very good to acquaint itself with the basics of the game, but not much more than this.

My theory as that bot players (at least those that use Monte Carlo methods) don't overplay, ever, and thus they have a very passive style. But as a beginner, you will play against other beginners, which means you will be straight at "Overplay Town". If you don't learn how to play against overplays and aggressive players, you'll never pass the beginner level.

Samura, might I suggest something? Since you have this program, set Igowin HD at its highest level and take however many stones would be not quite enough of a handicap. I think that might cause you to revise your opinions. I strongly doubt Igowin set to 15 kyu is using the MCTS algorith (it is extremely difficult to get this algorithm to work properly below a couple kyu; it becomes extremely erratic).

[Samura]

Oh, I didn't know that Igowin HD could not being using MC in the lower levels. I think this fact debunks my theory.

Anyway, one time I played an even game set as 3kyu, it was a pacific game where I lost by 11.5 points. I don't know if this helps my theory or not. Needs more experiments.

[Mike Novack]

Well I know that Many Faces (the program for running on normal strength computers) only uses MCTS for the top two levels and an AI based on "go knowledge" for all of the lower levels. The reason that this particular program, even when playing at the MCTS levels can give an explanation "why" a move was made is that it uses the "go knowledge" AI to produce a set of plausible moves (and then selects which is best using MCTS as the evaluator* instead of the AI evaluator). I don't know that it's the same for Igowin but I suspect that it is because weakened too much the MCTS algorithm would become erratic. Just the way probability works when the sample set is reduced too much.

Keep in mind that the MCTS algorithm evaluates moves based on probability of winning but ignores the margin of victory. That means if at some particular setting you lose by 10 points you might still lose by ten points taking one more stone handicap or two more stones, etc. In fact, observing this behavior is one of the ways you can tell MCTS. Try that 3 kyu level again this time taking a few handicap stones and see what happens.

* It can't give a "go reason" why the chosen move is better than the next best, etc. but it can give the "go reason" behind the chosen move by remembering why the chosen move was put into the "plausible move set" in the first place.