riadsala

The inevitable happened: AI can now play Go

Recommended Posts

Does KGS explain scoring when you finish a match? I guess what I'm getting at is that SmartGo has tuned in pretty well to my skill level and I find myself losing by 1/2 point sometimes, and I'd really like to see how it's calculated to understand which stones I could have contested to win that point.

 

 

Yes! you and your opponent need to click to mark the agreed dead stones and then it will tell you points from territory, prisoners and komi.

Share this post


Link to post
Share on other sites

If you enjoy go, it's mostly important to build up knowledge by playing and learning the really basic rules(like alphago) and then seek out a teacher when you have questions.

It's such a fun game, alphago is really bringing a lot of people back to the game as well as new players. Ive been playing over 20 years and I still learn new things all the time.

If KGS ever gets an html5 client out, we should start an idle go club. Since they released the source for KGS clients that should happen eventually.

Share this post


Link to post
Share on other sites

Apologies for slight derailment, but I've been thinking about the future of games for a good while now after the AlphaGo games. Especially interesting is the fact that a single instance AlphaGo performs relatively speaking only marginally worse than the clustered version.

 

a problem can be extremely difficulty and relatively easy at the same time
 
if a computer could play checkers in the 80s, chess in the 90s, and go in the 2010s, maybe it can finally play RTS games in 2050 and 4X games by 2100 :)

 
Do e-sport tournaments have requirements that the participants should be humans? While we are still far away from a system that could read the game image and control the game with a mouse and keyboard API, strategy-wise a system capable of defeating top humans in Lords Management/RTS could probably be built already.
 
The single player AI is a truly fascinating field though. The possibility of a game having relatively human-level understanding of the game state is really promising. It brings us a step closer to solving the problem of generating a system that can provide a satisfying gaming experience for the player.

 

Sometimes you just want to blob and steamroll. Sometimes you want a challenge. People look for different things from games, often without being fully cognisant of what they actually want. Varying the desired short and long term goals for the AI is probably very difficult to manage, as the information channel the system would have of the player's mental state is merely the in-game actions the player takes. But even ignoring all the psychological aspects, we already could do so much more with just some general understanding on the state of the game.

 

Hopefully this means that in the future the end game slog will be made both shorter and more epic. After the game reaches the point where the AI predicts that the player will win 80% of the games from this point onwards, it will change it's strategy fundamentally. For example, if the AI is aggressive, it will go for one final big push, sacrificing all else. If it's passive/pacifist it will just surrender. And perhaps a diplomatic AI faction will fracture, with some cities/armies surrendering and others joining some other AI faction.


I can also imagine some of the horrible things that will follow. Tutorials that will determine what kind of a challenge the current game offers. Ubisoft will have really scary player profiles. Oh, this guy quits the game if he fails a mission twice in a row in the same combat, but will keep trying as long as he get's a bit further every time. So we will always make the second encounter of the same enemy/jump really easy. And this guy is a masochist for collectibles, so we emphasize them on the minimap and advertise our collectible dlcs to him every time in Uplay.
 
Well, they probably do the latter already. Still, it's all really interesting, awesome and horrible at the same time. As is our future in general :-)

Share this post


Link to post
Share on other sites

That's interesting, and maybe deserves its' own topic. It would depend on what people decide the next challenge is.

Im not sure what comes after go.

I would so love to play against alphago. To think that computer assisted go training will be a thing!

I played against a guy online the other day and I think he was running a type of go bot because he made monte carlo type mistakes when he fell behind. (This was on OGS, where ranks get fuzzier around the dan level.)

Share this post


Link to post
Share on other sites

That's interesting, and maybe deserves its' own topic. It would depend on what people decide the next challenge is.

Im not sure what comes after go.

I would so love to play against alphago. To think that computer assisted go training will be a thing!

I played against a guy online the other day and I think he was running a type of go bot because he made monte carlo type mistakes when he fell behind. (This was on OGS, where ranks get fuzzier around the dan level.)

 

 

OGS is a pretty terrible place in terms of ranks and whatnot (it's a nice interface and I'm sure things will settle down in the future). It does allow for community groups to be made though, so we could set up an Idle club there anytime we like. Just now, I'm mainly playing on DGS.

 

As for coming across people running bots... this is pretty similar to playing against people who consult josekipedia and walthari's database during games. I've made my peace with it... my aim is to play well and winning or losing shouldn't really be a big deal. If some people decide to "cheat" to get a better rank, they're only really cheating themselves, as their game will suffer without their crutches.

I believe that AI is already more or less at the top level for poker (I'd expect a big showdown game to happen as soon as somebody stumps up a big enough prize pool.

Share this post


Link to post
Share on other sites

A year later and I finally decided to start playing, only done about 15 9x9 games and one 13x13 game. I'll get to the 19x19 board eventually!

 

Also, over new years AlphaGo apparently played ~60 online games against top pros. The American Go Association have started putting out commentaries on the games.

 

 

I enjoy this guy's videos as well

 

Share this post


Link to post
Share on other sites
On 12/03/2017 at 2:30 AM, eot said:

A year later and I finally decided to start playing, only done about 15 9x9 games and one 13x13 game. I'll get to the 19x19 board eventually!

 

 

 

 

My top tip is to not spend too long watching videos at this stage. The way dan players play is only good if you have the reading knowledge (and other things) to back it up. 

 

There's a a great saying for beginners: "lose your first 50games as quickly as possible" (ie, don't other think things at this stage, just play and you'll start to recognise patterns).

 

Good luck, and enjoy. If you'd ever like a game reviewed, let me know :)

Share this post


Link to post
Share on other sites

Nah, I know. Most of that stuff is way over my head, but I still enjoy them every now and they also have a motivational effect. So far I've been trying to review the games I play on my own (beginner mistakes are so obvious that even beginners see them in hindsight), but I've gotten two or three reviews from other people. I also decided to document my learning process with the game. Played my two first 19x19 games today. Surprisingly I won both!

Share this post


Link to post
Share on other sites
On 19/03/2017 at 9:35 PM, eot said:

Nah, I know. Most of that stuff is way over my head, but I still enjoy them every now and they also have a motivational effect. So far I've been trying to review the games I play on my own (beginner mistakes are so obvious that even beginners see them in hindsight), but I've gotten two or three reviews from other people. I also decided to document my learning process with the game. Played my two first 19x19 games today. Surprisingly I won both!

 

 

That's great! I really need to get into the habit of reviewing games and working a bit harder. I can play quite close to shodan on a good day, but I'm very inconsistent and often make stupid mistakes.

Share this post


Link to post
Share on other sites

Damn, shodan? That's impressive.

 

How long have you been playing for?

Share this post


Link to post
Share on other sites
On 26/03/2017 at 4:57 PM, eot said:

Damn, shodan? That's impressive.

 

How long have you been playing for?

 

 

note, I'm not actual dan level yet! On KGS, I bounce around 5/6 kyu. I've managed to beat a 4kyu in a competitive game (1hour clock time each). And while in Australia, nearly managed a win against a 3dan who suggested I was sandbagging. My play is too inconsistent to get up to 1dan on the net servers just now. I feel like I often get ahead in the first half of the game, and then stop concentrating!

 

I first discovered the game around 2002! But I only made an effort to learn how to play properly around 2010 I suppose. And being a full time researcher, I only have a limited amount of free time to spend studying the game properly.

 

I hear it's quite possible to go form 30kyu to 1 dan in a year. Probably easier if you're younger though!

 

 

 

 

Share this post


Link to post
Share on other sites
On 26/03/2017 at 6:25 PM, A Zen Master said:

On a go ai related note, a free program called Leela uses neural net tech also and is probably around 5-6dan.  It is also great for game and positional analysis.  You can put a position into it and ask it what it'd play.

 

https://www.sjeng.org/leela.html

 

 

I'm kind of sad to see actually. It reinforces the feeling I've had of more and more people turning to these tools while playing online. It could just be that I'm terrible at online play, but I've always found it odd that I'm closing in on 4kyu based on real board, competitive games, yet am stuck on 6-10kyu (depending on the server) while playing online. Lot of people using database search and josekipedia in the openning, and path analysis to cheat reading (that certainly happens on OGS, as I've asked opponents about it before).

 

I wish I had a local club to play real games. I guess I should make the effort to go to London one evening :)

 

 

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now