Video of Joseph Wilk's talk about Go

[PeterHB]

http://devslovebacon.com/conferences/ba ... lks/wei-qi

[PeterPeter]

A nice introduction.

Here is his KGS rating chart: http://www.gokgs.com/graphPage.jsp?user=josephwilk

[amnal]

It remains the one game that humans still easily crush machines.

It's interesting that this popular statement is becoming less and less true, and seems likely to totally cease being a selling point (to the extent that it ever has been) in the nearish future. Perhaps it will turn out to have been really important, and everyone will migrate to some large board arimaa variant

[palapiku]

It's interesting that this popular statement is becoming less and less true, and seems likely to totally cease being a selling point (to the extent that it ever has been) in the nearish future. Perhaps it will turn out to have been really important, and everyone will migrate to some large board arimaa variant

Such people will have to keep migrating every few years - at least, until we create an AI general enough that it beats humans at every game...

Although the perception that Go is impossible for computers will no doubt persist for a long time even after that. I keep seeing people refer to Go as the game where a beginner can crush the strongest programs!

[PeterPeter]

I keep seeing people refer to Go as the game where a beginner can crush the strongest programs!

I wouldn't be surprised if this puts a lot of early players off the game. GnuGo is no pushover on even its easiest level. I can imagine a lot of players play some 9x9, try against GnuGo on 19x19, get completely crushed, infer that they have no talent for the game, and don't bother again.

FWIW, I thought he understated how strong computers can be now, and overstated how easily a player will beat someone rated 1 or 2 stones below them.

[Inkwolf]

I keep seeing people refer to Go as the game where a beginner can crush the strongest programs!

I wouldn't be surprised if this puts a lot of early players off the game. GnuGo is no pushover on even its easiest level. I can imagine a lot of players play some 9x9, try against GnuGo on 19x19, get completely crushed, infer that they have no talent for the game, and don't bother again.

I don't know, I heard that statement when I first started, and was very frustrated playing Gnu Go, but I had the concept in my head that the computer was supposed to be easily beatable, and that I could defeat it if I learned a little more.

I agree that the improvement of Go programs seems more likely to put a beginner off, though. There seems no point to playing chess against the computer at all, knowing that programs can beat even the best players. It will be sad when Go reaches that same level, where you never get to believe that your own potential is greater than that of the machine. Especially considering how hard it is to find another human to play with without going online.

[PeterPeter]

There seems no point to playing chess against the computer at all, knowing that programs can beat even the best players.

I don't think that is an issue; you can easily choose a weaker engine, or hobble your current one to give you an even game. Something like Lucas Chess has over 100 different computer opponents to choose from, gradually increasing in strength from 'monkey' to one that can outplay any human. Then after the game you have the option of having a strong engine analyse it at a level far above any human, and suggest master-level alternatives for any of your moves.

The point is that early Go-players are likely to hear that computers are useless at Go from any number of different places; say, a well-meaning website that has not been updated for a few years, or an experienced player who does not keep up with computer Go. Why wouldn't they believe them? Then when they play against a computer, they lose badly, and think they are not cut out for this game.

[PaperTiger]

I'm going to be blunt. This talk was got so much wrong it was hard to listen to.

For computer go, the most recent record for a pro loss to a computer is only 4 handicap stones. The strongest bot on KGS is 6d, a level that already the vast majority of human players will not reach.

He talks about problems ending the game and agreement, but seems ignorant of the easy solution using area scoring. He claims computers have a hard time ending the game or counting, but a computer will happily win by 1 point, something that used to be a pro trick.

At 16:10 he shows a supposedly ended position where Black can capture a single stone on the boundary of White's territory.

[Javaness2]

Which bot is 6d?

[Uberdude]

Which bot is 6d?

Zen19D

[Boidhre]

There seems no point to playing chess against the computer at all, knowing that programs can beat even the best players.

There's an actually excellent use for a strong engine even for a pretty weak chess player. Set up a "won" position for Black, give the computer White and see if you can hold onto the advantage even though you're playing an engine far too strong for you to beat on even terms. Especially handy for endgame training. This is better than the converse of setting up a won position and giving a weak AI Black and trying to turn the tables because of the kinds of errors computers aren't very "natural" when the engine is heavily tuned down.

I suppose what happens is when the engine gets far stronger than the human it stops being an just an opponent and becomes a training tool.

[jlaire]

At 16:10 he shows a supposedly ended position where Black can capture a single stone on the boundary of White's territory.

There is that. And there is a move worth 40 points.

To be exact, he said the players didn't believe they can score any more points, and then something weird about the psychology of passing. Perhaps this is a very clever joke?

[jts]

Okay, people keep commenting on this, so I just want to share my opinion: don't watch it. Thanks to PeterHB for posting the link, but it's just a 10k giving a presentation on Go, with all the confusions, mistakes, and stammers you would expect therefrom. It doesn't reflect poorly on him at all, I'm surely plenty of 10k would have done a worse job; but no matter how long the discussion in this thread goes on for, do not click the link, it never gets interesting, you cannot get those minutes of your life back.

[Mef]

He talks about problems ending the game and agreement, but seems ignorant of the easy solution using area scoring. He claims computers have a hard time ending the game or counting, but a computer will happily win by 1 point, something that used to be a pro trick.

I'll admit i didn't watch the talk (since several here said to save the time), but for new comers (and relative to other games) ending the game is quite challenging regardless of ruleset. Go ends only when players agree the game is over...that's a little crazy when you think about it. Imagine if in chess your opponent kept making moves well after checkmate, or in backgammon if you kept rolling even though all of your opponents pieces were off of the board, simply because you didn't agree you had lost yet.

[Javaness2]

I'm interested in the background of the talk - can Peter tell us what the event was, or why Joseph was presenting the talk?