Page 1 of 2

xkcd

Posted: Wed Sep 11, 2013 9:01 am
by emeraldemon
Image

I'm not sure if there's anything particular triggering this comic, or if Randall is just now hearing about some of the computer go developments of the last few years. I do believe that computers will eventually reach professional level, but I think it's probably still between 5 and 50 years away.

Re: xkcd

Posted: Wed Sep 11, 2013 9:56 am
by Phoenix
You beat me to it. :mrgreen:

I'm wondering, too, if there's some new development I'm unaware of. I hate being unaware.

Re: xkcd

Posted: Wed Sep 11, 2013 10:47 am
by Mike Novack
Whether computers will get to that point probably depends on another conceptual breakthrough. It's not simply more powerful computers as brute force doesn't work well with go. Before the non-obvious MCTS method was developed go playing AIs were at about 6 kyu and they might be as good as 4 kyu now. The big, rapid advance was a conceptual breakthrough that with a little bit of help, "if following move A a higher percentage of random games is won than following move B then move A is better than move B" is enough to play a rather good game of go.

So now the strongest programs are playing about amateur 5 dan on powerful computers. Bumping that up to super computers might add a stone. The problem is that already into the region of diminishing returns for this approach as above a certain point the reliability of the percentages increases very slowly as the sample population is increased.

Another conceptual breakthrough is unpredictable. We may not have yet reached the limits of MCTS but I expect any serious improvement will be concpetual (some clever discovery about pruning, etc.).

BTW -- there is a great deal that computers aren't even close on. Try this, suppose you know 10,000 jokes. Somebody tells you a joke and you can select one from your list that is closely related in terms of "why is it funny". The difficulty with this problem is that one's total database of knowledge is involved. For example, the statement "the peas are ready to eat" and "the chickens are ready to eat" differ in the that the second is ambiguous and the first isn't in spite of no syntactic difference <<has to do with the properties of peas vs the properties of chickens -- note that we could add a third statement "the horses are ready to eat" and now whether seen as ambiguous has a strong cultural component>>

Re: xkcd

Posted: Wed Sep 11, 2013 10:52 am
by Codexus
Yeah, I think there are still a few years before computers really are stronger than the best humans at go.

However computers are now clearly pro strength at shogi, as proven by this year's denousen tournament but even there they still need to beat the current title holders. But it could happen soon.

I guess humans are not entirely obsolete yet. :mrgreen:

Re: xkcd

Posted: Wed Sep 11, 2013 11:14 am
by oren
Codexus wrote:Yeah, I think there are still a few years before computers really are stronger than the best humans at go.


Isn't a few years 'soon'? :)

Re: xkcd

Posted: Wed Sep 11, 2013 5:19 pm
by speedchase
I doubt it will happen in the next few years. Unless there is another huge conceptual breakthrough, we are really just waiting for MCTS to be able to outread pros, which won't happen anytime soon.

Re: xkcd

Posted: Thu Sep 12, 2013 10:08 pm
by Tim C Koppang
What about the underlying message of the comic? Regardless of when it happens, it seems clear that computers will someday outplay humans at Go. Does that mean Go players everywhere face the same crisis that hit the Chess world? Does it even matter except insofar as it means we humans have another studying tool? Isn't the point of Go to engage in a contest against another human, after all?

Re: xkcd

Posted: Thu Sep 12, 2013 10:51 pm
by Sampi
Damn, beat me to it as well :D

I totally agree with Tim's post, computers are just a tool whether or not they're better than us at whatever it is they do. What would be really scary about AI would be if they could actually reach a point of self consciosness!

Computers are better at checkers, chess, shogi, it will probably happen to go too. Computers couldn't have this discussion about how it sucks that computers are beating us at everything though ;-)

Re: xkcd

Posted: Thu Sep 12, 2013 11:04 pm
by karaklis
We shouldn't forget that after all it's human intelligence that makes the computer be able to beat us. So it's still a battle human vs. human, but on another level.

Re: xkcd

Posted: Thu Sep 12, 2013 11:59 pm
by wineandgolover
Tim C Koppang wrote:What about the underlying message of the comic?

The point of the comic isn't about go, which is just a benchmark against which to measure computer intelligence. It is about the rise if the machines. SkyNet is coming.

Re: xkcd

Posted: Fri Sep 13, 2013 12:56 am
by Kirby
Tim C Koppang wrote:What about the underlying message of the comic? Regardless of when it happens, it seems clear that computers will someday outplay humans at Go. Does that mean Go players everywhere face the same crisis that hit the Chess world? Does it even matter except insofar as it means we humans have another studying tool? Isn't the point of Go to engage in a contest against another human, after all?


I used to play chess a long time ago, but I'm not really privy to Internet chess competition... Do chess servers have issues with people using computers for assistance?

I suppose people could make use of joseki dictionaries, etc., already in go online, but maybe with strong computer go programs, we'll suddenly have many more 9d players online...? :-)


Edit: I guess a google search was called for before asking this question:

support.chess.com wrote:
How does Chess.com detect cheating?

•Part of our analysis involves comparing human moves to computer moves and looking at statistical significance.
•To remain effective, other aspects of our detection methods are confidential.



So maybe it's feasible for the go servers of the future to implement methods to detect cheating with computer pgorams.

Re: xkcd

Posted: Fri Sep 13, 2013 3:18 am
by Boidhre
Kirby wrote:
Tim C Koppang wrote:What about the underlying message of the comic? Regardless of when it happens, it seems clear that computers will someday outplay humans at Go. Does that mean Go players everywhere face the same crisis that hit the Chess world? Does it even matter except insofar as it means we humans have another studying tool? Isn't the point of Go to engage in a contest against another human, after all?


I used to play chess a long time ago, but I'm not really privy to Internet chess competition... Do chess servers have issues with people using computers for assistance?

I suppose people could make use of joseki dictionaries, etc., already in go online, but maybe with strong computer go programs, we'll suddenly have many more 9d players online...? :-)


Edit: I guess a google search was called for before asking this question:

support.chess.com wrote:
How does Chess.com detect cheating?

•Part of our analysis involves comparing human moves to computer moves and looking at statistical significance.
•To remain effective, other aspects of our detection methods are confidential.



So maybe it's feasible for the go servers of the future to implement methods to detect cheating with computer pgorams.


Computer aided cheating is a serious problem. There have even been instances of it in organised live tournament play. Some live chess servers have programs running on your PC that keep track of what processes are on the machine whilst you're logged into their server. Like MMOs do for catching bots. It's rather controversial.

Re: xkcd

Posted: Fri Sep 13, 2013 4:22 am
by Codexus
I play for fun at a level far below what could be considered competitive. If anybody wants to let a computer play in their stead to pretend that they are champions, I don't care. They are only hurting themselves.

Re: xkcd

Posted: Fri Sep 13, 2013 5:16 am
by hyperpape
Kirby wrote:So maybe it's feasible for the go servers of the future to implement methods to detect cheating with computer pgorams.
I think this relies on various chess engines being (mostly?) deterministic. If MCTS programs ever reach that point, you could compare a player's moves against the computer, but otherwise, it would be very tricky.

Re: xkcd

Posted: Fri Sep 13, 2013 6:01 am
by Mike Novack
Boidhre wrote:
Computer aided cheating is a serious problem. There have even been instances of it in organised live tournament play. Some live chess servers have programs running on your PC that keep track of what processes are on the machine whilst you're logged into their server. Like MMOs do for catching bots. It's rather controversial.


ROFLOL I bet "controversial". Pity the poor folks among us who would keep trying to convince the rest that is would be USELESS except to catch ignorant/naive cheaters.

You don't understand? Precisely how does ANY server know what is or is not running on that powerful second computer sitting next to me that has no connection to the outside world?