About playing bots

[msgreg]

Almost every time the issue of playing bots comes up, the discussion often goes to "playing bots teaches bad habits".

There is also the advice "play your first 100 games quickly." How do you reconcile this?

I've been able to play 200 9x9 games over the last month with Smart Go (Pro on the iPhone) and have finally unlocked level 13 (a month ago, I started at about level 7-8).

I think I have learned a tremendous amount in tactical "hand-to-hand", and which end-game plays are overreaching, and a little about being flexible in the beginning (i.e. don't pre-assume which corner(s) I'm going to take, in part: respond based on where my opponent plays looking at the whole board before playing). My number of "oops" plays are reduced (as is evident by the number of undos I execute). These seem to be lessons that will apply in future games with humans.

I still play regularly weekly in person 1-2 19x19 games per week, and I may have improved slightly against my regular opponent (who studies and plays much less). Perhaps about a half a stone...moving from me taking a 3 stone handicap to 2 stone handicap. On the side, I have recently started reading "Lessons in the Fundamentals of Go" at a slow pace.

Whenever I want to just win, I play black in igowin free 9x9 with a 2 stone handicap, in which I have a pretty standard play that nets an easy 16-35 points in my favor (it's limited to about 25k rating within igowin free). I can definitely see a difference in the styles of play at different ranks/computer strengths.

Overall, it's been a very enjoyable experience. I see the game differently even in the one month I've playing bots heavily.

I like that I can play and practice anywhere: waiting in line, at lunch, or even solving a few problems.

My focus has been: 1) have fun, and 2) learn more. My thought being that if it's not fun, I will stop learning.

So I just wanted to put out here that this is the way I'm learning, perhaps to encourage other people that were seeking a way to learn using bots, and to solicit opinions about how much I might be learning that I will have to unlearn later

To me, this feels like a balanced way to learn.

[badukJr]

Its because you'll end up playing positions you think are good, but are really bad that the bot doesn't know how to exploit. On the other hand, nobody can quickly show you why a position is bad.

The serious fact is, the best thing to do to improve is regularly play much stronger players over and over again, with a really short review at the end of each game. Finding a stronger player to do this is of course really tough, but taking the time to look is really worth it.

[msgreg]

The problem is, as you mention, trying to find players that are stronger that want to play. I also (strangely) do not prefer online play for 19x19 games. Also, I find that if I constantly play people stronger, it's less fun as the information I'm learning is coming faster than if I discover it slower, on my own with players ranked near me. It's often a case of learning together.

Again, my interest is in keeping it fun first, learning second. That's the only way go will remain a part of my life

I've already found the "learning bad habits" part when I switched from igowin 25k to smartgo level 7 (I have no idea how those relate in strength). I got creamed until I learned how to deal with the new tactics.

I guess the question is, as smartgo continually increases in strength as you beat it, am I again learning tactics that don't apply to humans? Further, wouldn't I learn those same (wrong) tactics if I play against players within a few stones of me.

Large shifts of perspective seem to occur rarely, perhaps 1-2 times between 30k and maybe 17k. It's these perspective shifts that seem to be the most useful.

What do you think of this rule of thumb: if a bot prevents you from having a perspective shift, then perhaps it is reinforcing bad behavior. If you still have perspective shifts with a bot, then you're on the right track.

[amnal]

I guess the question is, as smartgo continually increases in strength as you beat it, am I again learning tactics that don't apply to humans?

The answer is probably...it's plausible. But I'm not sure (possibly nobody is) how bad these tactics are or how signficantly they might damage your play if you start playing humans.

I also don't know what kind of engine smartgo has. Monte-carlo bots arguably are much less 'damaging' than the more traditional kind as they tend not to make the same bad move every time in a given situation. They still have a distinctive way of playing that perhaps you can learn to exploit, but it's much more adaptive and less easy to simply break.

Further, wouldn't I learn those same (wrong) tactics if I play against players within a few stones of me.

The idea is that each player would teach some wrong things, but every player would be a little different and have different bad moves or misconceptions (as well as good ones) so you get a lot more exposure to different kinds of play and its easier to pick up the good stuff.

This comes back to the MC vs traditional bot thing above. The real problem with traditional ones was that you could learn (even subconsciously) a moves that the bot would always respond badly to even though most humans would punish them. Even against a single human, the human will try different things each time and probably not always play the same bad way. Against a range of human players, they'll all have different influences all the time and play much more diversely and instructively. Again, there are interesting questions about whether MC bots play more comparably to this...though they still don't actually learn like a human does.

What do you think of this rule of thumb: if a bot prevents you from having a perspective shift, then perhaps it is reinforcing bad behavior. If you still have perspective shifts with a bot, then you're on the right track.

The perspective shifts thing seems a bit vague to me, since even if it's true different players will experience things differently. It's not clear to me that we can use perspective shifts as a useful measure.

[Unusedname]

I used to play against bots exclusively. I did undo's and everything.
I don't remember if I developed bad habits but what I do remember is how different it felt to play against a person

and also, for some reason the bot would always just abandon a group of stones resulting in me often going for a huge center.
which did work as a strategy against weaker players but quickly became impossible to continue doing.

And omg monkey jumps. Everywhere
Every single game.
I didn't even know why it was happening much less how to stop it.
I think that was the worst part about switching from bots to people.

But it doesn't seem like you'll have that problem of a transition phase since you play against both bots and people.
and like mentioned before, it really does help you get those 100 games out of the way.

[jts]

The main question is how you approach the bot. If you get really anxious about losing to a human, maybe bots are good for training wheels... Assuming losing to the bot doesn't make you anxious, too! But if you don't care at all about losing to a bot, and you're just clicking through the game mindlessly, that's no good either. There are people who come on KGS claiming they've been playing the computer for months and months, and they haven't figured out the first thing about go. I assume they can't bother to concentrate because if the computer wins, they'll just start again - no emotional investment, no improvement.

Not that improvement is the only reason to play.

I'm skeptical of the claim that it's hard for beginners to find stronger players to play. I've never seen a request for a teaching game go unanswered in the KTL.

[xed_over]

Again, my interest is in keeping it fun first, learning second. That's the only way go will remain a part of my life

Yep. just keep it fun.

http://www.usgo.org/news/2013/02/your-m ... -the-game/

[karaklis]

Monte-carlo bots arguably are much less 'damaging' than the more traditional kind as they tend not to make the same bad move every time in a given situation.

I think MC engines are even more dangerous. When you are playing such an engine and you are far in front or far behind, the engine will make unreasonable or at least suboptimal moves. When you are not strong enough to judge whether a move is appropriate or unreasonable, the probability is high that you adapt bad habits as well. A strong engine should crush me, clobber me, cream me until the board explodes, just like a 6d would crush me. Only in that case I would recommend to play an engine above DDK level.

[SoDesuNe]

http://www.usgo.org/news/2013/02/your-movereaders-write-why-its-better-to-fall-in-love-with-the-game/

Off-topic: Slowly but surely it gives me the creeps how many people think - and proclaim it as the thruth! - that they are limited by some inherent factor they cannot influence. What's the point? An ultimate excuse? You do not fail because your approach is wrong or you don't have enough free time to spend, no, it's because something that's out of your reach holds you down, so no problem, not your fault, just give it up or don't even try.
How can you improve in anything with this attitude? What's your take on life with this attitude?

If you are healthy, there is no stopping you: Go for it! What's wrong with that? That you most likely have to struggle? That you will fail often? That you temporarily have to face borders and pain because nothing comes easily? Come on, that's human! It's through challenges we grow.

Sorry, for the rant : ) Having fun is of course the most important part in anything but you determine how you have it and what's fun for you.

[Mike Novack]

This is a complex subject:

a) Some bots, even in the levels below where a MCTS algorithm is being used, allow an option "randomize play" in which case they will not always make the same move from a given position. The MCTS levels would always be random play unless the best move is obviously better than all alternatives.

b) The bad habits problem is worse if you have weakened the bot to your own level or below. As long as the bot is set to be a player several ranks above you it shouldn't be showing you the bad habits typical of your own level.

c) The MCTS programs aren't that adjustable for strength. This algorithm plays poorly if weakened too much. I believe there are theoretical solutions to making a machine opponent capable of playing weaker, say with probability X use the (weaker, more determinate non MCTS AI) for the next move and with 1-X use the MCTS engine with X being adjusted for the strength desired. Even a weak human opponent sometimes makes a very good move, more often if they are less weak.

[topazg]

When I play someone who tells me his move that I punished "Must have been good because bot X played it against me, or because it worked against bot X", I'll believe that bots are enforcing bad habits. Otherwise, I just don't buy it.

If I try something that works against Fuego, then it doesn't work against a human, and I can't find a sequence that gets me a better result against that human, I'll assume it doesn't work and stop playing it. I'm guessing most people work this way.

Bad habits are a case of blindly playing the same moves / sequences / responses (that aren't optimal) and not thinking about whether your move is best or not (this _is_ something I do), and this is where habitual thinking comes from. Bots do not have a monopoly over making humans do this.

If you want to play bots, play bots. If you start having games that are very fundamentally similar each time, find another bot or a human. Variety is key to getting stronger, your precise opponent's style less so. An opponent that plays better than you do is all you really need, bot or no. If you can find a stronger player that will review the game afterwards, that's much better, and you won't get that with AI (I don't think ALICE plays Go) - aside from that, play all the bots you want.

[skydyr]

Bad habits are a case of blindly playing the same moves / sequences / responses (that aren't optimal) and not thinking about whether your move is best or not (this _is_ something I do), and this is where habitual thinking comes from.

This demands a new thread: Joseki as bad habits.

[Amelia]

I'm skeptical of the claim that it's hard for beginners to find stronger players to play. I've never seen a request for a teaching game go unanswered in the KTL.

I've never had a problem to find a teaching game or a review, either. That's a very nice aspect of the go community, that teaching one another is so common. But when you play someone who is much stronger than you, you're asking a favor and not giving back much. It makes you feel a bit like a whiny kid asking for a ride on the motorbike. Also it puts a lot of pressure to play at your best, so as not to waste the time of the person who's teaching you. Which is good of course, but sometimes you just want a quick game between the washing up and putting the kids to bed.

But if you don't care at all about losing to a bot, and you're just clicking through the game mindlessly, that's no good either.

I do that sometimes. But bots are not very forgiving of mindless playing at least at my level. I don't think a beginner is in any danger of relaxing too much when playing GnuGo. Maybe it's different when you get stronger.

[skydyr]

I've never had a problem to find a teaching game or a review, either. That's a very nice aspect of the go community, that teaching one another is so common. But when you play someone who is much stronger than you, you're asking a favor and not giving back much.

I always got the impression that there was an expectation of reciprocity. You have been taught by someone stronger than you, so it is now your responsibility to teach those weaker than you. You may not be able to do much teaching right away, but the debt has been created for when you get stronger. In addition, teaching something is a great way to ensure that you understand it yourself, and to cement it in your mind, so it has a benefit for the stronger player as well.

[Amelia]

I've never had a problem to find a teaching game or a review, either. That's a very nice aspect of the go community, that teaching one another is so common. But when you play someone who is much stronger than you, you're asking a favor and not giving back much.

I always got the impression that there was an expectation of reciprocity. You have been taught by someone stronger than you, so it is now your responsibility to teach those weaker than you. You may not be able to do much teaching right away, but the debt has been created for when you get stronger. In addition, teaching something is a great way to ensure that you understand it yourself, and to cement it in your mind, so it has a benefit for the stronger player as well.

Sure. Rationally speaking, it all works out on a global scale. And I do look out for game requests like "Just learned the rules", so I can have the fun to explain what snapbacks and false eyes are. My point was just, there is much more pressure when you play someone much stronger than you. Psychologically, for me it's equivalent to a tournament game. And although it is very, very worthwhile, I don't like having that pressure for every single game I play.