I'm interested in buying MFOG12 and/or Crazy Stone 2013. I'm a beginner; just started this month. MFOGv12 18k trial seems just about perfect for me in strength (it rates me 19k).

However, people tell me that using AI is bad and will generate bad habits. I'm interested in MFOG and Crazy Stone as they appear to have educational features, like showing off territory. MFOG apparently plays "like a human" according to some posters, which sounds good to me. Crazy Stone has features to analyze good/bad plays and tell you what you did wrong, like a human player would.

Is there any current (2013) advice regarding beginners and AI? Does the "AI teaches bad habits" still apply to these AIs? Does it apply to DDKs? I'd imagine at dan level one might start to recognize patterns, but if an AI is many stones beyond me then certainly I won't be figuring it out, or I'd be that level, right?

Apart from the teaching features, I want AI for the general reasons of being able to pause/resume a game without annoying a human opponent.

Hi Milkman, the only people AFAIK who have no bad habits in Go are those very few who have good pro training from the beginning.
I don't know the percentage, but I'm going to guess a wild number of 0.01% of the Go population.
This means the 99+% of the rest of us will have bad habits in Go. Bad habits are like weeds -- they naturally grow,
anywhere and any time. If you want to have fewer bad habits, find a good teacher who actively
helps you and corrects your bad habits, just like a gardener does with weeds.

So, don't worry about it. Play whoever you want to play, and buy whatever software you enjoy.

I'm not familiar with the educational features of commercial Crazy Stone, how extensive those might be. With MFOG 12 you can ask it to explain "why" (why was a move made) and it can do this because the move will have been among those selected for so "go reason" (in other words, even when playing at the MCTS levels, that's just for the evaluation "out of this set of plausible moves, which is the best". But all the moves in that set were put there by the "go knowledge" based AI that is used for the lower levels of play).

Bad habits is a matter mainly a problem for the lower non MCTS levels. For that reason you shouldn't weaken a non-MCTS opponent enough to be able to play even games against it. But what has been said about humans is quite true. Your human opponents will also have bad habits. The problem in this case is when trying to learn by playing against a computer program opponent it will be the same bad habits. You would be just as bad off with a human opponent if you only had one. Maybe folks discussing this problem should be more clear, the problem isn't really human vs machine (with regard to bad habits) but always playing the same opponent with his/her/its set of bad habits. A different opponent would have different bad habits.

If you are a 22 kyu beginner you don't have to buy MFOG immediately. The free trial version plays at ~18 kyu (not MCTS) and the easy problem set should be about right. But even though this would slightly weaken the program, be sure to select the "randomize" play option (only refers to the non MCTS levels). Try it out, taking just enough handicap stones so that you lose much more than half the time (but win sometimes). Reduce the handicap as you improve. Once you are winning close to half the time playing even games you'll have to decide whether you want to pay for the program.

I've used Crazy Stone for awhile now and although it isn't like playing a person it does help to figure out mistakes in your play. If you keep it in normal mode there are also unlimited undo's so you can play out scenarios until you get it right. For example if you made a silly mistake and lost a group of stones you can go back and analyze why you lost them and then replay it to see if there was a way to save them. The territory feature also really helped me out in the beginning as even early in the game it shows how your stones are influencing potential territory. You can really start to see how the stones interact if you toggle back and forth while playing the game. Of course this isn't a replacement for studying tried and true Go theory. My advice is to get it to practice, but then get on one of the Go servers play some real people and have a blast.

All in all I wouldn't worry about it too much just mix it up.

First, if you play weak opponents, computer or human, you are likely to pick up bad habits from them. Since you are a beginner, set the level of the program so that it gives you at least 5 stones. Or play a human at least that strong.

Second, since computer programs do not learn across games, you may be able to discover weaknesses to exploit because they reply incorrectly to your bad play. Thus, that bad play is rewarded and reinforced. This problem is not so great with programs that incorporate randomness, such as Monte Carlo based programs, which I think all serious go programs do today. A trick play that works once against the program may well not work the next time. Anyway, if you find yourself playing to the weaknesses of the program, up its level or buy a stronger program. Or play against humans.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Thank you to all of you for the advice. I think I understand the point about AIs now and I won't worry about it too much. I intend to just supplement with them when I can't commit to a human game.

The thing to remember though, at your level can't really use a MCTS program as opponent. The algorithm can't easily be made to play weaker than a certain point without becoming very erratic, very unlike a normal human opponent of that strength.

Thus with MFOG, the top two levels are MCTS but below those the program is using its "go knowledge" AI not just for building the set of plausible moves but also to evaluate which of these is best (the higher levels use a MCTS based algorithm ofr the evaluation process).

When using those lower levels you'd want to to be sure to set the "randomize" option (makes no difference for the top two levels as the MCTS algorithm, based on probability, is always "random".

SOME of the AI programs may have "style" options. If so, just as selecting "randomize" will slightly weaken the program, modifying these from the presumably optimized settings (for highest average strength)would slightly weaken the program. But, although slightly weaker, perhaps better to learn from as not always playing the same way.