South Korean Go grandmaster Lee Sedol said Saturday that it was "not a defeat of humans" after his third consecutive loss in a historic five-game match with Google's computer program AlphaGo.
"It's just a defeat of Lee Sedol, but not a defeat of humans," Lee told a press conference after he lost three games in a row in the "century of the match," which was regarded as a representative fight between human brains and artificial intelligence (AI).
After more than four hours of play, Lee resigned as the computer program developed by Google's London-based AI arm DeepMind maintained an unassailable lead.
The 33-year-old grandmaster, who has won 18 world championships for 21 years of his professional career, said AlphaGo is an amazing program but has yet to rise to a level of God.
For sure, AlphaGo seemed to have weak points, which were shown in the first two games as well, Lee said.
Touching on why he was defeated for the third time, Lee said he lacked a capability of overcoming such a severe sense of pressure and burden he has never felt before.
Lee expressed his apology for the defeats in vain, saying that he showed powerless plays despite public hopes for his plays and wins.
In hindsight, it would have been difficult for him to win the first match as he had many misjudgments on AlphaGo's capability, while he missed many chances in the second match, which he played according to his strategy in an early phase, Lee said.
Though the final winner was already selected, AlphaGo's real capability could be evaluated more accurately in the fourth and fifth matches, Lee said, hoping for more attention to the remaining games.
The five-game match, which kicked off on Wednesday, will last until next Tuesday. The next match will be held on Sunday in Seoul.
http://news.xinhuanet.com/english/2016-03/12/c_135181873.htm
>>28198
It's interesting they describe it as just a 'computer program'. I guess IBM didn't need to build an entire Supercomputer to play chess.
>>28198
I want to see this computer program play Duel Monsters - I have the Heart of the Cards & the latest modern GOAT strategies - what does it have?!?
>>28200
Modern multi CPU systems allow to make computings in certain time with such complex algorithms like AlphaGo.
>>28198
Something I would like to see General Artificial Intelligence achieve is playing Quake III at at least amateur level using the same inputs and outputs that a human would receive, that is a massively difficult task.
>>28236
what? there are so many aim bots, it's super easy to make a program that completely owns every human player on the map...
>>28237
>the same inputs and outputs that a human would receive
Read with your eyes, not ass.
>>28243
aimbots do just send phony mouse input to the game
or do you want it to have a robotic arm that controls the mouse
>>28279
Also pattern recognition system.
>>28279
Yeah but aimbots read code and assess locations of objects via external code
Deepmind analyses pixels the same way our eyes do
>>28213
This would actually be interesting.
I'd like to see advanced AIs play games with certain random chance involved, also stuff like >>28236
If AI still becomes highly proficient in environments with lots of uncontrolled variables then it could start seeing military applications soon. Imagine an AI controlled drone tank where you don't have to worry it'll blow off some random civilian's house.
>>28279
>aimbots do just send phony mouse input to the game
no
>>28200
>I guess IBM didn't need to build an entire Supercomputer to play chess.
Deep Blue, Deep Thought, Watson, etc. are all usually described as "supercomputers" in the press. The blurb in the OP post has a lot of very awkward language, which doesn't surprise me given the source.
>>28236
they already have it playing above proficient skill levels in atari arcade games and 3D racing games with only the games pixels as it's input.... an FPS seems like it shouldn't be that much harder.
>>28236
Except, realistically, no.
Go is one of the most complex games on Earth, strategically, and requires far more than basic reflexes and pattern recognition.
You're cheating by using the "same inputs and outputs that a human would receive"; adding a mechanical challenge to the initial intellectual one.
>>28236
this is dumb as hell
it's easy to create AI that headshots you from across the map
you'd be essentially creating a handicapped AI, which most games already have
>>28474
>It's not that hard
>Except for that thing he said which makes it hard.
For fuck's sake, think before you post. Computer vision has been a part of AI for the last half century.
>>28474
Go has a far smaller branching factor than a high-framerate 3D game, the input data from a 640x480 RGB screen is also far more than a 19x19 black and white Go board(~20423 times more data), and is much more ambiguous, instead of Go's perfect portrayal of the entire game universe an FPS gives you a small 2D perspective of a complex 3D space, 3D objects usually look different from every angle so even knowing where anything(or what anything) is is a challenge. Quake is simply harder than Go, from a mathematical perspective.
>>28482
Aimbots receive perfect player coordinate information, a far cry from the hard to interpret massive 2d array of colour information that a human has to deal with when playing the game.
>>28299
An AI that can generally achieve arbitrary goals in uncertain and stochastic environments is the holy grail.
Perhaps our species ends like this:
>Make AI just intelligent enough to improve itself
>AI recursively improves itself
>AI efficiently converts all energy and matter in the universe into GOOG stock as fast as possible, because that was the goal it was given.
>>28213
Access to the shadow realm
>>28198
He says that now, but where will he be during the robot uprising. Once Alpha go teams up with /r9k/, were done.
HAHA
HE IS A PRO AT GO
GO PRO
I almost l o l'd
Haha
>>28686
>Once Alpha go teams up with /r9k/
That sounds kind of ironic.
>>28684
We better get our Millenium Items out...
AlphaGo has won match 5.
Lee Sedol: 1
AlphaGo: 4
>>28968
I'd like to see some more games with AlphaGo and top class pros.
AlphaGo was constantly making moves that went against the common conventions and winning, and moves that were so brilliant that not even the pros had thought them up over the years.
