eyeball_kid
Member
It seemed like (according to Redmond) that Sedol played a few safe moves mid-game, so perhaps he was too confident of his lead and AlphaGo's ability to respond.
-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
AlphaGo AI vs Go champ Lee Sedol (Game over, humanity is done)
- Thread starter Kinitari
- Start date
- Status
this is old news from after game 1. After game 3 he said it takes 5 9p in a group to beat ago... He is basically a wiz kid so you need to take his words with a grain of salt.
In 6 months after AlphaGo has had a chance to train a few hundred million games more, he's screwed...
Prototype Viktor
Member
but that would take out the best game of them all Game 2
No. Playing first is a large advantage. In Korea and Japan komi is 6.5 points.
Jack Remington
Banned
Honestly, fatigue is a pretty huge disadvantage for the human player. The computer can just crank out move after move for hours with no loss in processing power. It's tough for any human to sustain that.
Allowing humans to have a time advantage would even the playing field a lot.
Allowing humans to have a time advantage would even the playing field a lot.
who knows... even if this match happens it might take a while for it to happen.
It would...but it would also defeat the purpose because the time restrictions are an integral part of the game.
The rules are the same for the computer and the human. If the computer wins, it wins. That's fair.
Mac@NeoGAF
Banned
So the Brits have perfected AI and use a military satellite network called Skynet.
..just saying.
..just saying.
Duracelllll
Member
Well he did have the weight of all of humanity on his shoulders.lee looks so sad
Komi's value has been increasing steadily in the last century, it was not that long ago when it was just basically nothing, and players kept realizing how much it meant in the life of a game, so if anything, it will increase further, not decrease. Even with these games. AlphaGo did not win because of Komi. Outside of game5, which was very close, all the other games got won or lost because of battles in them, not because there was nothing more to be done and komi just got in the way, imho.
Edit: Wow at translation issues in the conference
Edit: Wow at translation issues in the conference
crazygambit
Member
It's disappointing that they never once got to the count in the end during the whole series of games. I really wanted to see that.
Otherwise I'm glad it wasn't a complete shutout it made it much more interesting and I would love to see how the N°1 player fares against AlphaGo.
Otherwise I'm glad it wasn't a complete shutout it made it much more interesting and I would love to see how the N°1 player fares against AlphaGo.
crazygambit
Member
This is actually a fantastic idea. Except for the fact that 2 people could start a game simultaneously with different colors and essentially force it to play itself thus guaranteeing 1 of them wins the jackpot.
It could work if AlphaGo only plays white. Lee Sedol said something about AlphaGo being stronger playing white because it doesn't have to overcome the komi.
Only one challenger at a time and what ever other rules we need to stop someone for cheating.
Is anyone seriously alleging that komi is a factor, here? I mean both players have played as white.Komi's value has been increasing steadily in the last century, it was not that long ago when it was just basically nothing, and players kept realizing how much it meant in the life of a game, so if anything, it will increase further, not decrease. Even with these games. AlphaGo did not win because of Komi. Outside of game5, which was very close, all the other games got won or lost because of battles in them, not because there was nothing more to be done and komi just got in the way, imho.
Edit: Wow at translation issues in the conference
crazygambit
Member
Well AlphaGo's only loss was as black. Lee Sedol also said the computer was stronger playing white. So, yes, it might be.
I understand game 5 was very close. Maybe a lower komi would have made a difference.
DragonSworne
Banned
Lower Komi is giving black a handicap lol.
Might as well give Lee Sedol an extra stone to start with.
crazygambit
Member
I'm talking about it being 7.5. I understand it's usually 6.5 in japanese/korean rules.
Whether that 1 point was enough to make a difference in the outcome of the game I have no clue, but LSD said many times that he felt it was more difficult (and meaningful) to win with black, maybe he knows a little bit about Go.
The last game was a lot closer. I think Lee took a few games to understand what it was like playing against AlphaGo. If it was a 10-game match or something along those lines, it would probably be a lot closer.
Not to mention playing one match every day is super tiring and a big disadvantage for the human. I never understood why that schedule was considered a good way to match a human against a computer. I thought that was rough on Kasparov when he played against Deep Blue, and now they did something similar with AlphaGo.
And I don't think Ke Jie is really talking trash. He gave himself a 60% chance to beat AlphaGo in its current form, which really isn't much, and acknowledges that it could be unbeatable by humans in a matter of months.
Not to mention playing one match every day is super tiring and a big disadvantage for the human. I never understood why that schedule was considered a good way to match a human against a computer. I thought that was rough on Kasparov when he played against Deep Blue, and now they did something similar with AlphaGo.
And I don't think Ke Jie is really talking trash. He gave himself a 60% chance to beat AlphaGo in its current form, which really isn't much, and acknowledges that it could be unbeatable by humans in a matter of months.
Aruarian Reflection
Chauffeur de la gdlk
AlphaGo currently ranked #2 in the world. It is genderless and nationless. Standing in its way is Ke Jie
http://www.goratings.org/
http://www.goratings.org/
eyeball_kid
Member
True, but then again the AlphaGo that played against Sedol is not a single computer. It uses many linked computers, but Sedol only has one computer, so you could say that's equally unfair. I would have preferred they used the single-computer version of AlphaGo.
crazygambit
Member
It wasn't really 1 game a day though. It was 5 games in 7 days IIRC, which I think is sort of standard for these kinds of matches.
Michael Redmond is listed as Japanese. Looks like we lost him at the latest racial draft.
Coriolanus
Banned
Huh. Somewhat bizarre that no women can break the top 100.
Prophet Steve
Member
Although I see how it makes for a more fun competition, I can see how they are using the distributed system if their goal is to find flaws in AlphaGo. It's less useful if they use a deliberately weaker version of it as some issues could just be solved with more computational power.
Acerac
Banned
I love that Ke Jie and Alphago hold the same W/L ratio against Lee.
crazygambit
Member
Not really if chess is anything to go by. Women are rated much lower there as well. As for why that is, there are many hypothesis.
I'm surprised he's rated so low. 541? Even Fan Hui is rated higher and he's only 2p. Redmond is supposed to be 9p, which you would think makes him much stronger than a 2p. Plus he seemed like a badass at the commentary.
Edit: According to that site he hasn't won a game in 2 years, so it might have something to do with that. But it seems super incomplete, since I clearly remember they said in the commentary that he was the first American with over 500 wins in tournament games and he had something like 580 today. The site only shows him at 44 wins since 1988 so it's definitely wrong.
Are there any official ratings for Go players?
From his wikipedia, it's seems like he's more about that commentator life now.
And ELO ratings are generally only calculated against other ranked players, so it's possible the site isn't including all of his wins smashing nobodies.
call_kotaku
Member
I don't think you can drop down in dan ranks, but I might be wrong
eyeball_kid
Member
What I mean is, the more computers they add to the AlphaGo system, the faster it can process moves thus giving it more and more of a time advantage. It would be like if they told Lee he could bring in 10 of his pro friends to analyze different variations on the board. The AI should be judged against its algorithms, and not gain advantage just because it has more brain horsepower.
Acerac
Banned
During the last match one of the devs from Alphago said that they had multiple iterations of the program running just in case one crashed.
It seemed less that they were there for collaboration, and more that they were trying to avoid just forfeiting the match were something to go wrong.
No one would have watched because Lee would have completely annihilated AlphaGo, that's way too big of an handicap.
call_kotaku
Member
But as far as I understand it a human brain still vastly out-thinks whatever network of super computers Google can manage to muster. Lee Sedol (and every other human) is still infinitely better than AlphaGo at basically everything except playing Go, you just can't specialize and focus a human brain to perform a task as effectively as you can with a computer.
The point is that human innovation and technological advances allow us to build a specialized tool that can solve a certain problem better than humans can, just like how we built machines that can lift heavier and travel faster than our bodies can.
crazygambit
Member
They did say tournament games, which I assume are against other professionals. So I'm thinking it's definitely incomplete.
Even if he's mostly a commentator now and doesn't really play anymore, that wouldn't be reflected on his ELO since it would remain unchanged (one of the big problems of ELO to begin with).
To really be fair, Lee Sedol should have to use the same algorithm AlphaGo uses to decide on moves. It's the only way.
cpp_is_king
Member
Uhh, the distributed version of AlphaGo only wins against the single version 70% of the time. It won against Lee Sedol 80% of the time. The sample size is small obviously, but that still suggests that Lee Sedol is about evenly matched with the single version.
I think the purpose of this exercise is not to evaluate whether one human brain is better at Go than one discrete computer, but rather whether it was possible for a computer in any form to beat a master human player. To keep things in perspective, before AlphaGo people thought we were still at least a decade away from this point. We know that at some point in the future we will have access to the same level of computational power as that distributed system, so view this more as a proof of what is possible.
This argument also is becoming more arbitrary as we move into a future where it is common to have access to cloud distributed systems for processing. If you are going to limit AlphaGo to a single computer, then why not also limit it to a single core? Or to an SoC? There is not a clear distinction of what we define as a 'computer' anymore.
1. I'm assuming that you don't need to be a ranked professional player to participate in all tournaments, although maybe I'm wrong.
2. I'm also assuming that you need to play all the time to keep your skills up, which is what I was implying by mentioning that he's switching his focus to commentating. If you click on his record, he's been on a losing streak lately.
- Status