Go Go 囲碁! Computers are stupid, but they don't judge

[inverarity]

I was not feeling well today, so I didn't go to the go club.

I have long avoided playing computer games because they are time sinks, and computer go proves to be as addictive as any other game. I am still not playing human players online as often as I should, but I am definitely improving against MFOG.


According to MFOG, I am now 9-10 kyu. I think this is an inflation of several stones, since I don't do nearly as well against human players at that rank. But I decided to test its ranking system.

Remember when I first started playing against MFOG, and even at the 18-kyu level (this is the level it plays at if you download the free version) it was often beating me? So, if I am 9 kyu now, I should be able to give it a 9-stone handicap and still win when it is playing at 18-kyu.

It turns out, I can!




This was a 9-stone handicap game (meaning I was white, and the computer got 9 stones to start). In the above game, you can see that at 18-kyu, the computer's limited look-ahead ability sometimes causes it to do very stupid things, e.g. that long line of black stones along the upper edge. Even a beginning human player will immediately see after putting down the first stone that there is no escape, but the computer played 9 before giving up. I won this game by 93.5 points. And it wasn't a fluke - I pretty consistently win by 70-80 points. The 18-kyu computer player that was so frustrating for me a few months ago is now a pathetically stupid opponent.

For ranked games, the computer now usually plays against me at the 9-kyu level, and I am winning about half the time. In the game below, I was black, the computer was MFOG (9 kyu), and I won by 10.5 points.



I know that computer go is a bad habit. I'm going to try to play more human players, honest.

Now, I am going to go do some writing tonight, I swear...

(And oh joy, my wrist has been bothering me lately and it really hurts tonight. I fear either excessive keyboard typing or jujutsu has damaged it.)

Comments

[kith-koby.livejournal.com]

I think this is true for most computer games. There appears to be a distinct lack of long-term planning.
There have only been two computer games I ever found challenging, and even these were only when set on a high level: Chess and epic strategy games like Total War or Crusader Kings II. The former is obvious; the latter may be my fault for lack of experience, or because of how complicated the game is, so that even I cannot really look ahead enough.

[tealterror0.livejournal.com]

Both of the games you cite are insanely complicated, so that probably has more to do with it. I think the problem is that it takes a shitton of work to make a good AI, and most game developers are lazy so they just give the computer cheap advantages instead.

[kith-koby.livejournal.com]

Yeah, I think it's a combination of both factors, but primarily how complicated the game is. The fact is, even with a great deal of experience, it's still insanely hard to survive/win in TW and CKII.
I have to wonder though, is it that Go is so much more complicated than Chess so it takes too much or effort to make a good AI, or is it simply that no one's bothered trying making a really good AI for Go.

[tealterror0.livejournal.com]

It's the former. Mostly because Go is a 19x19 board, while Chess is 8x8, so there are many many more possible game states.

[kith-koby.livejournal.com]

Well, that's what I thought, except that TW and CKII are just as complicated, if not more (a lot more). So it's not just that. It probably has to do with lack of funding for a good AI, something computer games have less of a problem with.

[tealterror0.livejournal.com]

That's why I said that TW and CKII are only difficult because the computer is given unfair advantages, something you can't do in Go outside the handicap.

And I would disagree that Go is less complicated than those games. It's a 19x19 board and you can place a stone anywhere there isn't already a stone. If you try to brute-force calculate variations (which is what the chess computers do) it gets really complicated really fast. I don't know that there as many options in the computer games.

Computer go

[inverarity.livejournal.com]

In fact, computer go has been extensively studied and is an active area of research for AI scientists.

Besides the depth problem that tealterror0 mentioned (just using brute-force look-ahead algorithms, go is exponentially more complex than chess), there is the fact that go is strategically more complex than chess as well. Heuristics for evaluating the value of a move in chess can use fairly well-defined metrics (material, positional advantage, etc.) based on the current state of the board. But in go, a stone played on one side of the board can have implications on the other side of the board 20 moves later, and any move might be good for profit but poor for influence, or vice versa, which means whether or not it's a good move depends on which strategy the player is pursuing.

[shinygobonkers.livejournal.com]

yay for tangible improvement :]