Computer AI


Why are certain AI’s available to some people while they aren’t to others? For example, I never see Aquabot when I try to play against a computer, but other people have played against it today.


Bots are run mostly by users like me at their own convenience. Sometimes you need a computer for something else.


I believe some high-level bots may be rank-restricted - meaning only players whose rank is within a certain range can challenge them.


FWIW - if anyone ever wants to play against a 9P version of Leela in a browser / outside of OGS, this is available here:

Side note: you can also vary the strength of the bot you’re playing against by hitting the [CHANGE] button in the upper right corner.


The bot owner can decide, if the bot should be listed in the challenge computer dialog. Some bots are not listed for various reasons.

Some bots only play games the maintainer had scheduled.


Hey, for anybody still reading this thread - I’m somewhere between 13-15kyu and I have found that online version of Leela to be a very useful learning tool.

I wanted to give a few caveats and describe how I use it. So, there are many high-level AI available that will provide the user with heat-maps of how the AI evaluates the board, and suggests best next moves. I’m thinking of Lizzie and Leela specifically:

Leela (the sjeng desktop version)
Lizzie (here’s a guide for those not conversant with github)

While both of these are great tools, I would say that they offer the most benefit to intermediate / advanced players, because their guidance is so high-level. Players in the 17kyu - 25kyu range will have a hard time absorbing that guidance or applying it in un-assisted games because those AI can juggle multiple priorities so much better than beginners can. They also play invasion moves or throw-in moves much more fearlessly because they can spot weaknesses and do 20-move look-ahead in ways that are years beyond a beginner’s capacity. They are a great window into how a high-dan views Go strategy, but may leave the beginner more confused and alienated than enlightened.

In order to really benefit from that guidance, I had to do my own leveling up in these specific areas

  • Being able to follow sente/gote, especially knowing when to break that order and ignore a gote move to save something small in exchange for making a different sente move that creates the potential for bigger profit somewhere else
  • Learning the stages of the game and the shifting focus on priorities between Opening, Early Midgame, Midgame, Endgame, and what is the most important element to attend to in each.
  • Putting names to specific strategies at each of those stages (i.e. “Create vs prevent large extensions” or “Peep and extend back towards thickness to prevent a potential moyo” etc), and being able to identify and compare different strategies for a given move to see which one was the more important one at that stage.

So - back to the leela-one-playout AI - I find this version particularly useful because the interface makes it easy to

  • hide and display the AI recommendation for best move
  • turn the AI auto-respond on and off.

So, this is how I use it as a teaching tool:

  • Start a non-handicapped game playing Black against the 9P version, knowing that I will be using the AI assistance, with auto-respond enabled
  • Once we are past the first 2-3 moves of the opening, I always take the time to do my own analysis and identify what I think are the 4 BEST MOVES ON THE BOARD
  • Once I’ve done that, I take a closer look and really think about the different priorities those moves represent - is it more important to secure the corner White just approached? Or can I do another sente move to approach a different corner, and still save the corner White approached with a clever joseki? Out of my 4 best moves, I pick one move that I think is the best.
  • Only then do I hit the [BEST] button on the page, and see what move Leela would do
  • If it’s the exact move I picked - I feel smart and my learning is re-enforced
  • If it’s one of the other 3 moves that I identified but didn’t choose, I take the time to think about why Leela’s choice was more efficient than my choice - what different strategy does it represent, and how can I use that going forward?
  • If it’s a completely different move than the 4 I identified, what strategy does that represent, and how did I miss that opportunity? How can I train myself to see those opportunities going forward?

As I go through the various stages of the game, I also take note of how the AI’s priorities may be different than mine. So, for instance

  • In the Opening, sente approach moves are so important that it is worth the risk of not immediately responding to approach moves, or even letting one corner get double-approached if it provides you with greater advantages elsewhere. This took me a long time to figure out, and watching how Leela juggled those priorities helped me gain the confidence to do it in un-assisted games.
  • In Early Midgame I find myself too focused on local fights, whereas Leela will tenuki as soon as the stones are close to being settled so that she can move out into the middle or threaten to cap/enclose weak groups. That was a big eye opener for me strategy-wise.
  • Once I’m fully into Midgame, having sente can leave me with Options Paralysis - I’ve got 3 different attacks I can make, and 2 different fires I need to put out. How can I choose which of those 5 things I need to do first? Finding my own top 4 moves and then seeing which one Leela picks has been very instrumental in helping me identify which of those priorities are most important.
  • Also, I find that (in Midgame) Leela will often pick my 2nd priority move and (after a gote response) THEN play my 1st priority move. It took me a long time to figure out that that this is also a good strategy in that you pick off the low-hanging fruit of moves where you can predict the single response (that don’t require a long fight), and then play the moves where a sente attack is necessary, but the response tree could go in many directions. The idea here is to string as many sente moves in a row before getting down to battle. It’s an Endgame strategy that works just as well in Midgame.
1 Like

I’m impressed by your level of analysis for your go level.

I mean i just discovered how i made incredible stupid tenuki in middle of joseki when i was like 10kyu around, so i didnt have the tools you have now, just a book could make me doubt and rethink (or a teacher).

1 Like

Yeah, I feel like my kyu rating is kind of all over the place. Part of that is I don’t play very often, and when I do, I play correspondence games that take weeks or months. I beat an 11kyu player a while back, but then realized that this was a fellow who mostly played blitz/live games and didn’t do a lot of reading, and our game was his first correspondence game he’d played, so my reading gave me an advantage.

Then I played a 12kyu and totally shot myself in the foot by trying to do too many things, and lost a game to a 14kyu by making a bad-shape blunder at move 43 that put me down by ~10 points and never recovered.

I find it much easier to study go than to play it because - playing against bots - there’s no pressure when I make mistakes and flub a game. Putting words around explaining what my 5 options are is much easier for me that picking which one of those 5 options is the most important move (esp. in Midgame).

I’m also one of those people who spreads their available creativity across multiple hobbies. So, right now, my natural brainstorming juices are flowing towards my music recording projects, and another fiction writing project I’m doing. Unfortunately, that means my desire to learn more Go is somewhat thin on resources. Eventually, I’ll get tired of the music and fiction projects, and circle back to playing Go, and writing my for-beginners articles (which I really should get back to someday.)

But yeah - I’ve leveled up to the point where I have a better understanding what I’m doing wrong. Now I have to do the hard work to pick a strategy, and commit to figuring out how to do it right.

I see, interesting How each Of us has his own way.
Maybe I could suggest some light minded Games with reading on the spot? Looks like you could get strong quickly

1 Like

Your vote of confidence is appreciated. If you’re into

  • slow correspondence
  • reading my lengthy comments evaluating the current position and
  • giving me notes on strategy as we’re playing

I’d be down for a teaching game if you’re into it. Cheers.

Just realized I should probably @ you to my reply if you’re not watching this thread.

I disagree with the focus on the ‘best engine move’. Lizzie sort of encourages this by making the move stand out a lot, but there are often a lot of reasonable moves even at the professional level.
In KaTrain I color them by point loss to show kyu players (like myself) which moves are fine to play even if they differ from the AI’s opinion, and which moves just lost you 25 points. Go is still a very open game with lots of possibilities!


I completely agree with that^. Part of the process for me in terms of finding my 4 top moves, is growing more aware of the different strategies each may represent.

There are times when one of those strategies is much more risky and aggressive than the other 3. Even though a 9P AI would be more comfortable taking that route, I find it much more difficult to follow as a DDK - and sometimes I consciously choose to ignore the AI’s “best move” and select the less aggressive strategy because it has other benefits with fewer risk.

At the end of the day, it comes down to me selecting the strategies that I feel most comfortable following based on my own level of play, and getting the AI assistance to work towards that goal, rather than just blindly copying a 9dan playing style.

1 Like

Hum no I am not now in that mood for lenghly Games with analyse and comments, Sorry.
I suggested you the opposite btw, go play more and live.