"What is the fastest way to improve?"

I respond: "Well nowadays they have bots that are pretty good, dan-level already. So instead of playing only other beginners who will expose you to tons of mistakes and bad moves, try and play as many games against dan-level bots as you can until you reach the bot's level. Play people around your skill level only when you feel bummed out after too many losses or you just want human interaction every once in a while, but focus on bots and getting those games reviewed by stronger players."

Would you consider this good or bad advice?

I'd say it's both good and bad advice,

It's good advice, since playing games against stronger players, and reviewing them, is a good way to improve quickly. But it's also likely to give you certain bad habits, but...

This is true of any focused play study play. If you play the same set of players over and over again, you develop habits and playing style based on what works against them. Playing a bot just gives you a certain set of weird ideas and habits.

These days, with fairly strong MC bots, I'd say they're a decent training tool, but that I'd hesitate to make them my only training tool.

Oh, I seriously doubt playing bots would give you very bad habits (if any) -- especially compared to the bad habits you'd get from playing other beginners.

Like anything, I think it's another training tool. I played against igowin for a couple thousand games before I really started playing any real people (besides one real-life friend who introduced me to the game).

Personally, I'd like to see a decent mix, but my study has never been all that focused, so guess I can't really say all that much.

_________________
KGS: schultz [?].

Bad habits are like weeds: they are normal and natural.
Without an active outside force (e.g. a gardener), bad habits naturally grow everywhere.
(This is true not only with Go but many other disciplines as well. )
Regular games and reviews with much better players are nice, of course,
but even 6-dan, 7-dan+ amateurs have bad habits themselves and will not be able point out all those of a beginner.
The only way (I know of) to nurture only good habits and weed out all bad habits from the start is
to study with a good-level pro from the very beginning.
This is one reason some pros can quickly tell whether or not someone has had pro training by looking at their moves.
This thread has a game of Iyama when he was eight: viewtopic.php?f=13&t=3873
One impressive feature is the two children's moves were all good habits and no bad habits.
But tuition is probably an issue for many, so bad habits run their natural course.

The reason I think this advice is bad is that the so-called strong bots are still playing a lot of bad moves. They are just the kind of bad moves that don't immediately lose the game at that level.

I don't know enough about how all bots work to make this claim, but it is also possible that one can develop a strategy that works well against a particular bot but which would not against most human opponents.

I may be wrong, though, and my comment applies mainly to 19x19 go. If you're the type that believes new players should play 9x9 a lot more than they usually do, then bots are very strong at that size board. That game is so tactical that it's hard to say you'd get bad habits from playing it.

When I get my butt kicked by Zen on 9x9, it seems I am not learning anything from Zen. There is nothing I can emulate or take into my own 19x19 games. But what I do see is the vast holes in my own perception, and the degree to which I am capable of fooling myself.

Hmm... Strange... Then why am I able to beat Leela at 9x9 giving 2 stones, sometimes even at 3?

If I had to guess, that's probably because I played a lot against bots in 9x9 at the beginning. First against Igowin, then against Leela. I didn't really learn a lot from that, except how to beat the bots. Against real players, I'm a few stones weaker at 9x9 than at 19x19


To answer the original question:
I have only one data point, that is, myself. When I started playing seriously, the bots were about 5k. Playing against them did get me to about 17k on KGS, even though I was able to play even against the bots. Only when I started to play against humans, I was able to improve further, and then I could suddenly give handicap to the bots.

That experience leads me to believe playing bots exclusively is a bad idea. And I think MC would make that effect even worse. Of course that doesn't mean you shouldn't play against bots at all

I don't think it would hurt, but my caveat would be this: playing against bots all the time, you are unlikely to encounter many of the most common and elementary sequences and patterns that arise in human play. It would really be imperative that the student is concurrently studying some strategy books and books on basic invasion, enclosure, or reduction sequences.

I think getting a beginner to play bots is good but only because it's a good way to play lots of games very quickly. For this reason I always recommend they play as many games as possible against igowin until they get to the point where they can beat it in an even game. I'm a firm believer of the "lose your first x games as quickly as possible" proverb for whatever x you feel is appropriate. Sometimes I think x should be just 100 sometimes I think I haven't reached x yet.

I'd also recommend igowin over a stronger fast playing easy to use 9x9 bot. I think it is strong enough that you can learn basic shapes (hane/connect on the first line type stuff) so that the games are close to `real go'. And weak enough that it provides a reasonable target to actually be able to beat (eventually).

Only playing bots until dan level seems so wrong though. I'm not too worried about picking up bad habits (I think this applies to anyone who plays only a small set of players and doesn't have a good, strong teacher to step in) but I think there are some other valid arguments against it.
1) Human interaction: for most I don't think "every once in a while" is enough to maintain enjoyment of the game. Besides, it's much more fulfilling to get to a reduced handicap against a long-term playing partner than a bot.
2) I think a review of an evenly matched game can be more valuable than a review of a handicap or unevenly matched game and I know I'd much rather review a game between two humans (even if one is not there to benefit from the review).
3) Bots are rubbish at reviewing.

I'm a firm believer too, but I prefer "Finish 100 games as quickly as possible." Not crazy about the 'losing' part.

Perhaps my bad advice vote is wrong. Strictly speaking, playing strong bots might not be a bad way to improve, although I like Ed's post and you need to be able to deal with all manner of strangeness, and playing people of various levels coming up provides that experience. Of course, going over your games with stronger players is solid advice.

But I react strongly against a go world where we only play other people when we "are bummed" or on those (rare?) occaisions we "might want human interaction once in a while". A go world where everyone followed this advice until they reached dan level would be, I suspect, a sad one indeed. Highly bummed people, beaten down by bots, in a bad mood and reacting, probably, badly when things go wrong and behaving like the worst examples of KGS behavior because they have not played humans enough to learn how to act, and then mostly when the are predisposed to react badly.

My best guess is that you want to (mostly) be playing opponents whom you could beat at a nine stone handicap, and a bot that a beginner could beat at a nine stone handicap would be playing in an idiosyncratic way.

Probably everyone benefits from occasionally playing a much-stronger player and getting slaughtered, but if you get slaughtered at every game, you're unlikely to see how the different parts of the game fit together. To get a sense of how the opening flows into the middle game, and the middle game flows into the endgame, you need to win on the board once in a while.

Besides, it's not clear to me that the premise (you pick up more bad habits from beginners than from strong players) is necessarily true. A beginner can take a move from a pro game and turn it into a bad habit. It might be more accurate to say "if you play the same beginner again and again, you'll pick up his bad habits."

Hmm. Well, I guess it's not learning from you

Thanks for the data point. I lean more toward the "bad advice" response now. Another thing about minimax algorithms is that if you are emulating the kinds of moves they make when they are winning, you are emulating (or worse, unsuccessfully attempting to emulate) some serious slackness. It may not be best way to improve for a human to try to find the highest probability 0.5 point win.

Minimax can be viewed as a style. It's not wrong, and if you have the resources to pull it off, it's theoretically the best style if your goal is to win the most games that are right in front of you today. But I don't think it's the best way to learn.

At that last U.S. Congress, I saw a lecture where a 3 dan showed a position like this. The pro said that " is fine if you want to be 3 dan for the rest of your life." I think his point was that if you're not going to approach white's corner and at least allow some potential for complications, you're not getting exposed to the kinds of challenges that will help you improve.

Click Here To Show Diagram Code

[go]$$B
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . 1 . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Are there any bots with a tutor mode, where they warn you if you make a move they think is bad? Or just suggesting the best moves for each turn after the game is over? Gnubg (an open-source backgammon bot stronger than any human) has those features and they're very useful.

I think the advice is OK in the sense that playing bots will probably help you, but I don't know if I necessarily relate "fastest way to improve" with "play bots". Playing bots might be fine, but it seems that other methods of study might be more effective toward improving.

_________________
be immersed

We need to perform some human experiments. Who will volunteer to be the Tarzan of the Go world, raised by computer go-bots until they can beat them in even games and then surprise attack the go world?

I believe this is bad advice. It still would be, even if the bots were professional level. For: The most important precondition to get strong is to keep playing. If you learn Go as just another internet addiction you will likely drop out sooner or later.

So true. To make an analogy, there is a type of strength training called SuperSlow. Proponents claim it's very efficient (something like 20 minutes a day twice a week) but the problem is that almost everyone quits. For the 1% or so who can continue, it may be great, but it's hard to recommend in general.