-
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
call_kotaku
Member
the commentary makes it seem like Lee is playing the same opening he did in game 2 but with a twist, so I hope that means he figured out something and actually has a plan
A Grizzly Bear
Member
I believe it's because they're playing Chinese rules, which uses a 19 x 19 board.
killer rin
Member
They were talking about it earlier. Something about Alphago being trained on a 19x19 board.
We could, but by making them fight each other for our entertainment, we are only giving them reason to rebel
Games like Star Craft wouldn't interest me. I mean, a computer with perfect timing and dexterity would probably cream a human on those factors alone.
Unless we're talking a robot that is a camera pointed at a monitor, and an arm moving a mouse, then I can see it being somewhat interesting, but FPS' are unbeatable with a perfect AI aimbot/headshotting everyone instantly. You would have to program Star Craft *down* to make an AI competitive. Go is good because it is purely strategic.
Unless we're talking a robot that is a camera pointed at a monitor, and an arm moving a mouse, then I can see it being somewhat interesting, but FPS' are unbeatable with a perfect AI aimbot/headshotting everyone instantly. You would have to program Star Craft *down* to make an AI competitive. Go is good because it is purely strategic.
19 x 19 is the standard goban. smaller 9x9 boards and 13x13 boards are often used with beginners, since the 19x19 board can be overwhelming.
um, spoiler alert for the ending of WarGames lol
http://senseis.xmp.net/?19x19Board
Balance of local and global play, plus maybe some spiritual numeric connotations.
cpp_is_king
Member
Why is a chess board 8x8? That's the size.
has to be odd otherwise it wouldn't be symmetric. If you're wondering, 9x9 and 13x13 also exist, but mostly for teaching
StrategyFan
Member
the jump from go to brood war is quite large
it's really hard to make a good starcraft AI, even with unlimited APM
still, i don't really like APM-based games for AI. something turn-based like prismata or a card game would be more interesting imo
what does it mean to beat a human who possibly had better strategy because of superior micro or because of an inability to get tired? it doesn't make a strong case for anything in particular, even though it's a really hard thing to do.
do we have to start calling the brood war elites "bad players" if they can't sustain high APM in the end game for a long enough time? most of them will be growing past the prime age for hyper-attention games
starcraft 2 would be better candidate. not because of the game quality, but simply because the APM requirements are lower and more people may be actively dominant at it
I wrote a brood war AI years ago.
getting this far only took a few days of work. the native AI is way worse than what others have created
writing an robot that doesn't know anything about starcraft and "discovers" how to play it well by trial & error is a separate story though
but i don't really think it means anything for game-playing AI to write a "player" that doesn't even know the rules of the game. that type of problem is more suited for general learning systems that have nothing to do with being good at anything in particular
this would be nice
I plan to make something like this happen (not for civilization, but a different game), but there's very little prior work to learn from or look at to get an understand of what is even possible or realistic. every game in the genre just has a cheating AI
i think that playing against 99% of players without cheating is doable
being able to challenge the top 1% is extremely hard, and I have no idea how long it will take to get there.
it's really hard to make a good starcraft AI, even with unlimited APM
still, i don't really like APM-based games for AI. something turn-based like prismata or a card game would be more interesting imo
what does it mean to beat a human who possibly had better strategy because of superior micro or because of an inability to get tired? it doesn't make a strong case for anything in particular, even though it's a really hard thing to do.
do we have to start calling the brood war elites "bad players" if they can't sustain high APM in the end game for a long enough time? most of them will be growing past the prime age for hyper-attention games
starcraft 2 would be better candidate. not because of the game quality, but simply because the APM requirements are lower and more people may be actively dominant at it
I wrote a brood war AI years ago.
getting this far only took a few days of work. the native AI is way worse than what others have created
writing an robot that doesn't know anything about starcraft and "discovers" how to play it well by trial & error is a separate story though
but i don't really think it means anything for game-playing AI to write a "player" that doesn't even know the rules of the game. that type of problem is more suited for general learning systems that have nothing to do with being good at anything in particular
this would be nice
I plan to make something like this happen (not for civilization, but a different game), but there's very little prior work to learn from or look at to get an understand of what is even possible or realistic. every game in the genre just has a cheating AI
i think that playing against 99% of players without cheating is doable
being able to challenge the top 1% is extremely hard, and I have no idea how long it will take to get there.
They to tested alpha go against itself already.
They had a distributed and a standalone version playing against each other. The distributed version won 70 percent of the time and is the version facing sedol
Well, the size of boards for Chess variants is determined by the number of pieces, but Go's rules scale to almost any size.
Before AI I want to see Starcraft/RTS evolved to Multi-Input. Right now we have one cursor and many hotkeys that generally result in a large series of sequential commands. I'd like to see the mouse replaced with some sort of touch interface to give players the ability to routinely issue a much larger number of parallel commands. Instead of Maynarding a spread players would be controlling a small number of units simultaneously.
Looks like fun. Can't wait to get my headset.
This stream was pretty good yesterday. Should start in 10 minutes or so.
He's thinking a lot before making a move.
call_kotaku
Member
considering that he apparently lost all 3 games within the first 50 moves, maybe taking as much time as he can isn't a bad idea
https://youtu.be/6hROM_bxZ9E
There are some cool "15 minute recaps" of the matches by the commentators.
There are some cool "15 minute recaps" of the matches by the commentators.
Silphonica
Banned
alphago's weakness is betago.
smh haha
smh haha
Well. It's good to make use of your time to think. If the clock hits zero, he'll go into overtime. He'll have only 1 minute per move in that situation. He'll have 3 allowances, where if he goes past the minute on a single move, it'll reset, using one of those allowances. It definitely makes things more tense at that point, though.
That's the time he needs to formulate a strategy though. What he is doing the past 3 games clearly doesn't work.
Jack Remington
Banned
Rooting for Lee harder than I root for my hometown teams in the playoffs.
Won't matter much if he doesn't have time to put it in play
Kaiser aus Ubermensch
Banned
This is amazing!
I cant believe an AI can compete with humans in terms of judgement.
The impact all these technologies will have on human application and how we'll structure our societies within the future are bound to be huge.
It's been a while since I've played go but can someone break down for me how to claim territory? Its connect 4 but is it diagonally as a whole or directly?
I cant wrap my head around the rules as they're detailed on Wikipedia's page.
I cant believe an AI can compete with humans in terms of judgement.
The impact all these technologies will have on human application and how we'll structure our societies within the future are bound to be huge.
It's been a while since I've played go but can someone break down for me how to claim territory? Its connect 4 but is it diagonally as a whole or directly?
I cant wrap my head around the rules as they're detailed on Wikipedia's page.
Maybe, he will probably still lose, but he is attempting something different to improve his game. His previous approach will earn a definite loss.
He is probably stung by the fact that yesterday's game was lost early on by the AI securing a huge territory at the bottom of the board.
diagonally or in a straight line it doesn't matter. Diagonal groups are more vulnerable to attack because they have cutting points, but for the purpose of territory, it is the same.
He took 15minutes for his last move.
Prototype Viktor
Member
he's getting worse and worse as these games go on with how much time he takes game 3 AlphaGo had 50 minutes when he had 12 now AlphaGo has over 70 minutes
killer rin
Member
Man, AlphaGo really wants that center territory this game
Natural, since he realizes how strong an opponent he is facing.
Funnily enough, there is an episode where a guy proclaims that computers will do so.
Garou
Member
No, the original implication if I remember correctly was that by playing online against "unlimited" players will do that. Not related to AI.
Oh, it's been a long time since I watched the show, probably remembering it wrong.
Yeah but he's overthinking very move he makes, so he will have no choice but to rush his moves now, and that IMO will very costly in the end game.
Jack Remington
Banned
Odds of Lee winning based on the current situation?
It's looking grim.
I think he's referring to this:
https://www.youtube.com/watch?v=RRscgUv4t4c&feature=youtu.be&t=16m45s
And yeah, this one random guy says that computers will be better than humans.
(Also while we are on the topic of HnG, this parody made me laugh https://www.youtube.com/watch?v=ma0d4glXkM8&feature=youtu.be)
- Status