Hello all,

I was wondering what sort of preferences players on Life in 19x19 have when they play.

Do you only play real people, mix it up with AI games or tend to play more with AIs?

Have you observed differences in moves or even play style?

If your opponent's identity was masked and you could not communicate with them, would there be a way you can tell if they are human or an AI?

I watched the AlphaGo documentary for the first time the other day and noted that Go commentators described Alpha Go as a playing just like a human, albeit an aggressive one. I find this very interesting. Not necessarily limited to AlphaGo, but has anyone experienced a time when they felt their opponent AI played with human characteristics? (hopefully my English makes sense...)

I started learning to play Go a month ago and have only played against two AIs so far - GoDroid and Cosumi. GoDroid lets me put adjust the strength of the AI, so perhaps this affects the way it plays. But I am too inexperienced to see any difference between GoDroid and Cosumi.

Sorry if this topic has been answered before. I couldn't find it when I searched old threads.Thanks for taking the time to read and answer.

EDIT: I did a Google site search instead and I can see that this question has been addressed by others but framed differently. Some posts point out one obvious flaw that bots make strange moves that humans never would.

This post reflects my limitations more than it reflects common reality, likely.

I'd like to play humans, in real boards. I'm trying to get to the point where I can play more or less focused for about half an hour. Otherwise, I feel playing live against a human is wasting that person's time. From time to time, I can play mail (the reason I joined OGS back when) with enough regularity to risk annoying humans.

Then, a matter of preference: I'd like to play against humans. I've never liked the feel of robots, although I'll grant that AI bots are a completely different beast, and have some nice moves. But playing only against bots is like playing fencing against a properly designed robot. Would the robot be better? With enough image recognition and without limiters in the engines, indeed. But if we go that way, we might as well give it a machine gun and throw the whole idea out the window. Tennis players (I think baseball, too) use robots to throw balls; you could build a robot that threw balls so fast no player would dare to stand in their way: 6-0, 6-0, 6-0, Top ATP in a season. What's the fun in that?

I feel Go is a... Dô [道], I guess. There might be a point in trying to be better than someone else, there's no point in trying to be better than a machine. There's a great point in trying to be better than oneself. I can't recall what Chinese pro said, after AlphaGo, "maybe now we can see what Go is about". I quite agree. The illusion of the race to the top, and rank and all that has been proven an illusion.

If and when we can configure AI to follow specific styles (besides having a catalogue of weights and their strengths; I'd like a "menu", or even real people styles emulation), I think that'll be a great tool for improvement. But it will be a tool.

YMMV, of course. Take care.

_________________
一碁一会

Hm, I will have to look at some trick moves later. At my level, I would not be able to tell if a move was a trick or just a plain old move!

Cheers.


Sorry - not quite following. Do you mean you played the moves by email?

Yeah, I see what you mean with the tennis analogy and AI being a tool in the end. I feel that is what all technology should be seen as. The human factor, whether it is performing arts or sport, we're drawn to it because there are emotions, thinking, long hours of practice and skill involved.

Cheers.

I try to only play real people. As others have pointed out, online go servers are now frequented by bot-enhanced players and according to Baduk Docter and Antti, too, this is a mayor problem when trying to get a match at the highest level.

On KGS I haven't had a bad experience so far. But I use automatch, long-ish time settings and I am just Shodan, so I guess it helps if you're in the non-swimmer's pool rather than the open sea.

When facing an opponent around my rank, I guess I can tell, if they are using a bot. That being said, they also could be sandbagging really hard and the moves would be equally confusing to me.

The most obvious difference between a bot and a human would be probing and tenuki to me.

_________________
My "guide" to become stronger in Go

Strong AIs like LeelaZero are hard to distinguish from human play, especially since humans have been imitating AIs. The main difference is that LeelaZero makes fewer mistakes.

About weaker AIs: I used to play against cosumi when I was a beginner. The AI is not bad for a beginner but its style doesn't change much so it can become boring. The first year I found it so strong! Now I think it makes strange moves, it doesn't have a good sense of which groups are weak.

I just played it again.



Don't take my moves as a model of good play (I am kyu level). At some point I started to play slack moves because I was leading by a large amount, and I failed to kill the bottom group although it shouldn't have lived. Move 17 was a misclick, although the move was OK.

The AI doesn't realize that the O7 group is weak.

Move 82 makes no sense, a human player above beginner level wouldn't waste time saving 3 unimportant stones like that.

Move 96 is strange, a human player would continue the attack on the two black stones.

Sorry, I got that phrasing back when to refer to turn-based servers and I haven't been able to get rid of it. OGS still calls it "Correspondence".

Sorry if I caused any confusion.

Take care.

_________________
一碁一会

Thanks for playing and posting. That was very useful as I have actually not observed/replayed any full matches on a 19x19 board. Some moves (both yours and Cosumi) didn't make sense to me but I think I got the gist about Cosumi not recognising weak groups well.