> I heard somewhere that playing against a Go software program can teach one proper playing habits.

The conventional wisdom is the opposite of this, on the basis that they predictably play the same set of bad moves that you'll tend not to see refutations for - unlike human players, who may be equally bad, but will be so in a range of interesting ways, each of which teaches you about the strengths and weaknesses of the others.

This is probably less true for modern monte-carlo bots, but probably applies somewhat to gnugo - especially if it's pushed into a 10 minute game.

On the other hand, you didn't actually say what your problem was, or why you don't like playing it, or why these games were not 'serious'.

Yup.

I once came across someone who might have been DDK in reality, but had created a dan-level account for himself by giving Gnugo bots 6 stones and playing essentially the same game over and over and over and over, exploiting the persistent misplays that are built into that engine.

Playing those bots does not teach you proper playing habits.

Actually, MC bots tends to be fairly crappy teachers unless you are looking to understand the value of tactical fighting and basic group safety. The problem is that MC bots will play more non-optimal moves when they feel they are ahead or behind (at least if the bot in question isn't using some kind of adeptive komi in its calculations). And that makes it quite difficult to learn from their play.

Traditional min-max bots are much better for learning in my opinion, especially for newer players. These bots have more consistent strengths and weaknesses that make them far easier to learn from. Yes, you'll pick up a few bad habits when it comes to areas where the bot is weak, but that is really not very different from playing players of a specific rank and server.

Also, using appropriate handicaps to make the fight more balanced will alleviate the problem. The biggest complaint I see about playing bots is examples of someone learning to exploit a weakness in the bot and then using that repeatedly without adjusting handicap to fix the problem.

Although if you are exploiting a weakness in that way, adjusting handicap will generally put you into a bind when the handicap is enough for the bot to prevent the exploit, because without the exploit to lean on, the players has little of use.

Dishonest play is essentially the single biggest hinder to learning. And that goes for playing both humans and bots.



And this basically sums up both of your games. You are trying to win by fooling your opponent (which works great against weak double kyu players). Those double kyu tricks won't work against GnuGo however.

For example, when you cut at K4 in the first game, GnuGo probably knows that your stone can't escape because it has read it out. You on the other hand keep trying to break it out repeatedly until you eventually give white a massive bottom as well as the right bottom black group.



If you want serious games then you should probably take your opponent seriously. (including accepting handicaps against a stronger player)

You are the one playing dishonestly trying to probe for weaknesses in GnuGos play. GnuGo on the other hand is a very honest go engine (in fact, that could be said to be its strongest weakness) that responds honestly to your attempts to fool it.

Using a conventional style of play, I've been able to win against mfgo09kyu a few times.

A GunGo3 series program has very strong reading ability. I estimate its strength to be 3-5 kyu. If it is meant to imitate a human Go player, than perhaps I should play against it more. Otherwise, I should stick strictly to playing human opponents.

As for accepting handicaps from the program, I have not been able to get any rated games from the GunGo engines on KGS. (Is it possible to adjust handicaps when challenging GunGo to a game?) Against human players I can get handicaps, but almost everyone there avoids playing weaker opponents. (On IGS, I get appropriate handicaps whether the game is free or rated.)

A general rule of thumb: play the bot, so long as it is still kicking your ass. As soon as you start beating it, adjust handicap until this isn't the case any more. As soon as you find yourself making a move because you know how the bot will sub-optimally respond, change to a new bot.


Actually, most of that could apply to humans as well.