It is currently Fri Mar 29, 2024 8:37 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 16 posts ] 
Author Message
Offline
 Post subject: Provocative Question
Post #1 Posted: Thu Mar 22, 2012 7:37 am 
Beginner

Posts: 18
Location: Germany
Liked others: 5
Was liked: 2
KGS: GoIngo
Hello,

in this month, computer go has made several big steps:

* the wins of Zen on 19x19 against Takemiya Masaki (9p)
at 5 and 4 handicap stones

* the win of Pachi2 in the KGS bot tournament at slow time controlls

* the impressive performance of Zen13 in many free games on KGS

******************************************************

Sometimes I like to ask provocative questions.
Here come a few. What do you think?

* When will be the first time that a human pro takes handicap from a
bot on 19x19? (Giving the bot White and komi=0.5 is counting as handicap.)

* When will be the first time that a human pro loses a game against a
bot on 19x19, when the human took handicap?

* When will be the first time that one of the strongest humans takes
handicap from a bot on 19x19?

* When will the first time that one of the top human player loses against
a bot on 19x19, when the human took handicap?

* Will it ever happen that a top human player loses against a bot in a
19x19-game where the human got 4 handicap stones?

I know that there are other hard steps which have to be mastered before
by the programmers. But nevertheless...

Ingo.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #2 Posted: Thu Mar 22, 2012 8:11 am 
Tengen

Posts: 4380
Location: North Carolina
Liked others: 499
Was liked: 733
Rank: AGA 3k
GD Posts: 65
OGS: Hyperpape 4k
As I expressed elsewhere, I have trouble understanding how MCTS could play at the level of top professionals. What's your take on that?

_________________
Occupy Babel!

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #3 Posted: Thu Mar 22, 2012 8:46 am 
Gosei
User avatar

Posts: 1585
Location: Barcelona, Spain (GMT+1)
Liked others: 577
Was liked: 298
Rank: KGS 5k
KGS: RBerenguel
Tygem: rberenguel
Wbaduk: JohnKeats
Kaya handle: RBerenguel
Online playing schedule: KGS on Saturday I use to be online, but I can be if needed from 20-23 GMT+1
Ingo Althofer wrote:
Hello,

in this month, computer go has made several big steps:

* the wins of Zen on 19x19 against Takemiya Masaki (9p)
at 5 and 4 handicap stones

* the win of Pachi2 in the KGS bot tournament at slow time controlls

* the impressive performance of Zen13 in many free games on KGS

******************************************************

Sometimes I like to ask provocative questions.
Here come a few. What do you think?

* When will be the first time that a human pro takes handicap from a
bot on 19x19? (Giving the bot White and komi=0.5 is counting as handicap.)
I guess it's still ~5 years ahead, at least
Quote:
* When will be the first time that a human pro loses a game against a
bot on 19x19, when the human took handicap?
6-7 years ahead. As soon as this is needed to keep even, a pro-slip can happen
Quote:
* When will be the first time that one of the strongest humans takes
handicap from a bot on 19x19?
Real handicap, like black with two stones? Maybe in 10 years
Quote:
* When will the first time that one of the top human player loses against
a bot on 19x19, when the human took handicap?
Probably never
Quote:
* Will it ever happen that a top human player loses against a bot in a
19x19-game where the human got 4 handicap stones?
No. I think.
Quote:
I know that there are other hard steps which have to be mastered before
by the programmers. But nevertheless...

Ingo.

_________________
Geek of all trades, master of none: the motto for my blog mostlymaths.net

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #4 Posted: Thu Mar 22, 2012 8:57 am 
Gosei
User avatar

Posts: 1435
Location: California
Liked others: 53
Was liked: 171
Rank: Out of practice
GD Posts: 1104
KGS: fwiffo
I wonder if man-vs-machine matches will disappear when go bots start getting better than top humans, the way they have with chess. I don't know much about chess but my understanding is that current chess engines would completely crush top pros in a straight-up match. They tend to win even win with one hand tied behind their back (e.g. no endgame or opening tables) while giving heavy odds like pawn and a move. Although elo ratings from automated computer-only matches are not directly comparable to human elo ratings, I think that it's fair to say that the chess machines are leagues stronger than the best humans.

It's been suggested that top pros would need to take three or four stones to play God, but given the experience from chess, I wonder if that estimate might be low.

_________________
KGS 4 kyu - Game Archive - Keyboard Otaku

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #5 Posted: Thu Mar 22, 2012 9:54 am 
Tengen
User avatar

Posts: 4511
Location: Chatteris, UK
Liked others: 1589
Was liked: 656
Rank: Nebulous
GD Posts: 918
KGS: topazg
fwiffo wrote:
I wonder if man-vs-machine matches will disappear when go bots start getting better than top humans, the way they have with chess. I don't know much about chess but my understanding is that current chess engines would completely crush top pros in a straight-up match. They tend to win even win with one hand tied behind their back (e.g. no endgame or opening tables) while giving heavy odds like pawn and a move. Although elo ratings from automated computer-only matches are not directly comparable to human elo ratings, I think that it's fair to say that the chess machines are leagues stronger than the best humans.

It's been suggested that top pros would need to take three or four stones to play God, but given the experience from chess, I wonder if that estimate might be low.


This is very true, but it hasn't stopped chess engines being developed - it still seems as active as it's ever been in the last 10 years.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #6 Posted: Thu Mar 22, 2012 10:51 am 
Lives in sente

Posts: 727
Liked others: 44
Was liked: 218
GD Posts: 10
Top programs will reach 9D in KGS with normal or slow time setting before 2020.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #7 Posted: Thu Mar 22, 2012 1:50 pm 
Beginner

Posts: 18
Location: Germany
Liked others: 5
Was liked: 2
KGS: GoIngo
hyperpape wrote:
As I expressed elsewhere, I have trouble understanding how MCTS could play at the level of top professionals. What's your take on that?


Two and also still one year ago several MCTS programmers
had expressed their opinions that MCTS would run against
some threshold rather soon. I - simply - believed them.

Now I am unsure what to expect.
The heavens for computer go seem to be open.

Ingo.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #8 Posted: Thu Mar 22, 2012 1:54 pm 
Beginner

Posts: 18
Location: Germany
Liked others: 5
Was liked: 2
KGS: GoIngo
Hello Ruben,

nice to meet you again. We had played in a little 9x9
go tournament 3.5 years ago on Little Golem (and you
beat me).


RBerenguel wrote:
>> * When will be the first time that a human pro loses a game against a
>> bot on 19x19, when the human took handicap?

6-7 years ahead. As soon as this is needed to keep even, a pro-slip can happen


That will be indeed another interesting question.
Will top humans "simply" avoid playing bots when
bots become too strong.

Ingo.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #9 Posted: Thu Mar 22, 2012 1:59 pm 
Beginner

Posts: 18
Location: Germany
Liked others: 5
Was liked: 2
KGS: GoIngo
fwiffo wrote:
... I don't know much about chess but my understanding is that current
chess engines would completely crush top pros in a straight-up match.
They tend to win even win with one hand tied behind their back (e.g.
no endgame or opening tables) while giving heavy odds like pawn and a move.


That happened in a few experiments with chess bot Rybka, but
human chess masters might learn to do better. Unfortunately,
there is no real culture of handicap game in chess.

Quote:
It's been suggested that top pros would need to take three or four stones
to play God, but given the experience from chess, I wonder if that
estimate might be low.


Indeed, (future) matches where go bots give handicaps to top humans
may provide lower bounds for the number of handicap stones God might
give to humans.

Ingo.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #10 Posted: Thu Mar 22, 2012 2:00 pm 
Beginner

Posts: 18
Location: Germany
Liked others: 5
Was liked: 2
KGS: GoIngo
Hello pookpooi,

pookpooi wrote:
Top programs will reach 9D in KGS with normal or
slow time setting before 2020.


A question out of curiosity:
Is KGS 9D assumed to be pro level?

Ingo.

Top
 Profile  
 
Offline
 Post subject:
Post #11 Posted: Thu Mar 22, 2012 2:45 pm 
Honinbo
User avatar

Posts: 8859
Location: Santa Barbara, CA
Liked others: 349
Was liked: 2076
GD Posts: 312
Ingo Althofer wrote:
A question out of curiosity:
Is KGS 9D assumed to be pro level?
No.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #12 Posted: Fri Mar 23, 2012 1:29 am 
Lives in gote
User avatar

Posts: 655
Location: Czechia
Liked others: 29
Was liked: 41
Rank: 1d KGS
KGS: Laman
Ingo Althofer wrote:
Hello,

in this month, computer go has made several big steps:

* the wins of Zen on 19x19 against Takemiya Masaki (9p)
at 5 and 4 handicap stones

* the win of Pachi2 in the KGS bot tournament at slow time controlls

* the impressive performance of Zen13 in many free games on KGS

******************************************************

Sometimes I like to ask provocative questions.
Here come a few. What do you think?

...

I know that there are other hard steps which have to be mastered before
by the programmers. But nevertheless...

Ingo.


as you said, there are some lesser challenges to be beaten before the time for your questions comes.

when will be the first time that a human pro plays an even game against bot on 19x19? and when will the bot win such a game for the first time?

i would put both answers close to each other, let say around a year apart, and i estimate it might happen in 10 years from now. then a human receiving handicap could follow again relatively shortly afterwards, like 2 years, but i can't really imagine going up to your suggested 4 stones

_________________
Spilling gasoline feels good.

I might be wrong, but probably not.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #13 Posted: Fri Mar 23, 2012 4:15 am 
Tengen

Posts: 4380
Location: North Carolina
Liked others: 499
Was liked: 733
Rank: AGA 3k
GD Posts: 65
OGS: Hyperpape 4k
"Is KGS 9d a rough ruler for pro strength?": http://www.lifein19x19.com/forum/viewto ... =10&t=4398. The short answer is that it's not quite reliable, but it's very very close.

_________________
Occupy Babel!

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #14 Posted: Fri Mar 23, 2012 7:22 am 
Gosei

Posts: 1625
Liked others: 542
Was liked: 450
Rank: senior player
GD Posts: 1000
For programs to play at solid pro strength (say 5p or higher) I think a new approach will be necessary. The reason is that even the Monte Carlo based programs don't really "understand" go. A good position evaluation function still has not been developed. The chess programs with opening and endgame dictionaries built in take advantage of the relatively small size of the chess universe so they can play well with a somewhat crude position evaluation function.

It is vaguely interesting that Takemiya lost those handicap games to Zen but I'd like to see what would happen over a longer, multigame match. Personally I am not very interested in how strong programs are or could become. For me there is some pleasure in regarding go as a sequence of puzzles to be solved, which is how playing against a program feels, but playing against a human partner is much more. The old nickname for go, "shudan" (hand talk), sums it up; a game is a conversation and a mutual appreciation of the marvels of the game. This mutuality is simply impossible when my "opponent" is a computer program. I might be interested in using a pro strength program as a training tool as I think some chess pros do with strong chess programs. But a good "teaching" program would have to understand how humans can think about the game to explain mistakes in a useful way. Rather than simply saying move x is better than move y it would have to explain why move x is better, not just say the probability of winning is higher with x than with y.

People have moaned and groaned about such issues as defending the human spirit against the machine and held up go as a game that will never be "mastered" by programs but this is actually irrelevant. No one complains that robot driven automobiles have demeaned the human spirit because they can beat human runners in 400 meter races. Maybe the best results for go will come from "hybridization" of human brains and computers, hardwiring computers into brains to provide assistance to the brain, as I expect we will see before too long with computer assistance to overcome sensory defecits or physical impairments such as paraplegia.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #15 Posted: Fri Mar 23, 2012 7:42 am 
Lives with ko

Posts: 199
Liked others: 6
Was liked: 55
Rank: KGS 3 kyu
gowan wrote:
For programs to play at solid pro strength (say 5p or higher) I think a new approach will be necessary. The reason is that even the Monte Carlo based programs don't really "understand" go.


Why do you believe that computers need to understand go to play it any better?
Computers are not humans. They don't need to understand go, nor to rationalize about moves as we do. They only need to pick the best moves, that's all.

Top
 Profile  
 
Offline
 Post subject: Re: Provocative Question
Post #16 Posted: Fri Mar 23, 2012 8:12 am 
Tengen

Posts: 4380
Location: North Carolina
Liked others: 499
Was liked: 733
Rank: AGA 3k
GD Posts: 65
OGS: Hyperpape 4k
gowan wrote:
But a good "teaching" program would have to understand how humans can think about the game to explain mistakes in a useful way. Rather than simply saying move x is better than move y it would have to explain why move x is better, not just say the probability of winning is higher with x than with y.
Yes and no. It's true that a program that understands "this position is thick" is the best thing that's imaginable. But a program that can find points where your move leads to a sharp drop in evaluation is quite helpful even without that possibility. Similarly, if you think a particular line refutes a move, it can show you why it doesn't.

I can also imagine a MCTS program where you say "what about aji here?" and it finds lines where the opponent attacks and shows you refutations. Doing that well would be almost as much a problem of UI as programming. As time goes on, expect more of that sort of thing to be developed, but right now, the race for strength is so pressing that it'll take time for those ideas to be implemented.

_________________
Occupy Babel!

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 16 posts ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: Bing [Bot] and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group