The computer that mastered Go
Вставка
- Опубліковано 26 січ 2016
- Go is an ancient Chinese board game, often viewed as the game computers could never play. Now researchers from Google-owned company DeepMind have proven the naysayers wrong, creating an artificial intelligence - called AlphaGo - which has beaten a professional Go player for the first time. In this Nature Video, we go behind the scenes to learn about the game, the programme and what this means for the future of AI.
Read the original research paper at www.nature.com/nature/journal...
Sign up for the Nature Briefing: An essential round-up of science news, opinion and analysis, free in your inbox every weekday. go.nature.com/371OcVF - Наука та технологія
This game should be played with M&Ms or some other candy. You can then eat opponent's candy when you capture it.
+Filip Voska It has been done. You can do this with chess too: weiqi-to-go.net/2015/02/26/chess-and-go-tourney-pays-in-candy/
+Filip Voska Then half the professional go players will be diabetic
+Steadystrike
That is the chess community's evil plan.
Most ideal candy would have to be Reese's Pieces. They only have 3 colors.
Steadystrike Nah. Professional Go players hardly get to capture at all. Their opponents are just to good to waste moves that don't do much. It's when you have the 20 kyus play the 30 kyus you get the entire board captured.
"it was stronger than perhaps we were expecting." @5:00
that is not something I want an AI developer to say, ever.
+Isaac Garcia (dzdkidd6) speechless
+Isaac Garcia (dzdkidd6) so true, lol
+tapu nima The programmer guy with short hair seems quite modest and in a different level even though he is way above most academic professors.
No one programmed AlphaGo.
It programmed itself.
It's a Neuronal-Network.
That's why it won in the end.
Just laughed my ass off, this is so scary and true. It's only a matter of time...
Put the computers against each other to see the greatest go match ever.
If the AI are same than the result will be draw upto 94.3%.
@@sjzz There's no draw in Go if you play by common rules.
In chess, the use of clocks are pretty standard. I don't see any being used for Go.
How long does a game of Go last for typically?
+ChessNetwork They do use clocks in tournaments and the game then will last 1-6 hours. How ever some professional games will be even longer than 16 hours and those games are the spread out over several days.
+ChessNetwork They do use clocks. There are various time variants, blitz games can take around 30 minutes, while tournament games can take up to 3 hours.
+McRoos Ahh. Okay thanks. I can see why it'd take so long. 6ish hours is a ballpark duration for a tournament chess game.
hi Jerry!
+alivfishland Did you like the squid?
Kind of disappointed that we didn't hear about this until after the fact, and that they chose to go the traditional way with a Nature paper instead of sharing every little advance immediately at arxiv as seems to be the (happy!) norm in ML these days.
But very, very excited to read more about their approach, and a BIG congratulations is in order!
Alright computers, come take over the world already.
+Ozymandias I wish i could live long enough to be destroyed by a computer.
+RAIN LOL
+Ozymandias they already have... how many people do you think could make it through a day without their smart phones?
Quantum Uncertainty Workshop He means AI, phones are useless without a human.
i was being cynical...
I've been following this news for awhile now. I'm glad to see there's some progress in this area. Almost two decades ago Deep Blue defeated GM Gary Kasparov and more recently Komodo 24 just beat GM Nakamuru. It was estimated that some people would be able to develop AI to defeat humans at this game right about now and here it is. I'll definitely be following the March match.
I, for one, welcome our computer overlords.
+Jeffery Liggett ken jennings after the watson challenge
+Jeffery Liggett Google Overlords*
+Jeffery Liggett You are just trying to brown-nose before the apocalypse. It won't work.
Here here!!!
i have mastered go fish
you have 6 of diamonds?
yaaas
+Tabulus Rasa No, go fish. (hehe, he'll never find out.)
Well I guess computers can now collect $200. Because they have...
Passed Go
Ha, Ha, Ha, Ha, Ha,Ha, Ha, Ha, Ha, Ha, at least 15 of us got your joke!
Brilliant Joke. Well done
Absolutely disgusting
Ooof
but can it beat hikaru ?
Walwalkn Wewnrkl
you are correct :)
+Afro ThuGz Maybe there is a ghost in the machine... the ghost of an ancient Go master.
Anders Öhlund
it will first need a new firmware to support the ghost
+Walwalkn Wewnrkl The AI did what Sai could only dream of doing, play the divine move!
+Afro ThuGz You mean Sai.
Who is Sai anyway?
lolol
Added to our favourites playlist.
Very good video. Good work, proud of you. (:
Congratulations to the team!
"It was stronger than perhaps we were expecting" - Not really what you want to hear when dealing with AI.
+neon32 This, exactly.
So it begins..
i learned about this game from a manga called hikaru no go. i learned a bit about go but the story was still interesting and fun without knowing much about the game
what I'm wondering is, did alphaGo learn to play from pixels like the neural nets that played the atari games? IMO, it would be great if did and if only one neural network was used without the search procedures. Good progress though.
If you had two of these machines playing each other would it be the same game every time? Or is there a level of randomness to the choices it makes even when it looks 20 moves into the future? I would really be interested in in seeing what would happen.
+Noah Brink-Goodman
Well, theoretically the losing computer (and possibly the winning one as well) would learn from their mistakes while playing, meaning they'd make different choices the next time.
Does depend on whether the creators and operators of these programs decide to turn the "learning-feature" on during the matches.
Great question!
This thought blew my mind.
Oddly, no, it doesn't for the past 6 years the 'best' AI have continually been upping each other.
Was a komi value used to balance to final score? Also what color was the AI playing? Regardless, this is a great improvement!
Amazing!
When will this become public? I want to practice go without setting up with a board.
It even just beat Lee Sedol 4 games out of 5. Couldn't believe it.
Is there anyone doing the brute force way? I mean that would be interesting thing to see as we say it will takes too long of time but if we let it run thousand years we get estimation how long that big of task would really take especially if we upgraded the software and hardware . . . I wonder if someone has already started this project?
found out abought go 20 minutes ago love it
are there any good documentaries on the history of 'Go' ? I want to see more about the masters of Go throughout history.
Alphago is the best documentary
So awesome!
"We're quietly confident that who knows what will happen."
I never considered that the holy grail of artificial intelligence would be so close to attainment in my lifetime.
What a fascinating development! It is very telling of how hard people are working to develop the field of AI.
I cannot wait to see the results of the match in March.
Is it possible that this (type of) program could beat Stockfish in the near future or is that still way off?
the conclusion statement is so true
"It was stronger than perhaps we were expecting"...
I imagine ko fights would be one of the more difficult things to program. How did they figure that out?
If Honinbō Shūsai sees this computer, he would be amazed by human's technology and progress.
Anywhere where I can watch cpu vs cpu/AI vs AI? Like games online.
Who did the music for this video?
I did not see any credits.
Where are the 5 Games against Fan Hui?
+David Ongaro here they are. www.usgo.org/news/2016/01/alphago-beats-pro-5-0-in-major-ai-advance/
+Bob Hearn Thanks, appreciate it
+David Ongaro Here you can replay all 5 games online:
www.go-baduk-weiqi.de/google-schlaegt-go-profi/
Wow; a Quantum Leap in computer go...
Humanities famous last words: "It was stronger than we were expecting."
Awesome! :D
They should get it ranked to see how many Dan AlphaGo has =)
where can I see kifu of this game?
Well, all the strategies described apply to computer chess as well. The thing is that Go needs an even stronger focus on heuristics compared to brute force but you can't play chess with pure brute force either.
That self-atari at 6:56...
Cool. a game that you play with Mentos and M&M's...................
+SlipKnotRicky
Impossible. Every time I play with those pieces, they mysteriously disappear.
Cabhan Listis Yep, same here....
+SlipKnotRicky Hahaha, yeah. I was thinking Junior Mints for the dark ones.
+SlipKnotRicky Then you'll having someone trying to get 1000 brown M&M's in the middle of the night, and that almost always ends up with someone getting beat to death with their own shoes. True story.
+SlipKnotRicky what about reversi/othello?
Incidentaly, the AI was also possessed with the spirit of an ancient Go master who wanted to play the perfect game.
I hope that Lee Sedol doesn't succumb to the pressure like Kasparov did against Deep Blue. I also hope that they give him access to some recent version of the program prior to the match; chess computers tend to have certain exploitable weaknesses, and you don't play against them like you play against humans... it would be interesting to see if a top go player with sufficient time can come up with something similar, or if the complexity of the game stymies such approaches.
Can't believe I just watched the promotional video for Skynet and kinda amazed by it...
I'll be back.
I don't think so if you use tensors in R`^n
We won't ever be absole to beat Sai !!!
I've only mastered counting Go:-(
I agree :-)
I am a house champion of GO fish.
Reminds me of the game "squares" i played in school.
Dr. Demis Hassabis, you played Go a little bit often, right??
I've heard of go. is the stragery harder than chess. really. I'll have to find out. buy a set and learn..is there a play the computer app. download etc. live by myself so regular player isn't an option really...
+kevin swithenbank loner
what happens when someone accidently slightly hits the board with an elbow?
5:33 GSV quietly confident
Just nitpicking: the objective is not to control more than 50% of the board, but to control more than your opponent does. For example, Black controls 45% and White control 40% => Black wins. E.g. seki is more usual than one would think.
+Nam Thai Think it depends on the ruleset.
+Jesus Christ Not really. Seki is a standard situation no matter what kind of Go you are playing.
+lollerskatez1 Stop saying 'not really'. A simple 'no' will suffice.
Stop making orders. A simple suggestion would suffice :)
lollerskatez1 I hurt your feelings.
And don't say, "Not really" like I know you are going to.
So, what about those GO computer games AIs? They certainly don't tree search each move till the end game. What are those programs about?
+Baichuan Ren Most recent ones use a monte carlo based tree search I think, or something like that. Apparently this one does too, but have an upper layer to reduce the tree beforehand.
+Richard Rondu Interesting, I will look into it, thx btw.
+Baichuan Ren You're welcome ;-)
i was wondering what material stones they were using, they look pretty :0
Probably Yunzi stones made in China.
How about shrinking the size of the board? That would reduce the amount of computations needed. Then just scale it up. The size of the board appears to be arbitrary. The strategy should be the same if the board is smaller.
This is like straight out of a scifi movie
Go is the game of black and white mentos, yummy
the winner keeps all the mentos
if we forced allowed... computer win with no draw and lose .can they build an AI playing chess? in fact I dont think any human program or self learning AI can do it its more difficult than Alpha go (playing chess without a draw and win most game)
@@smfanqingwu1474 Weqi has more possible variations than there are atoms in the universe a brute force approach to go would be impossible even with all the computing power on Earth combined.
can alpha go transfer it's experenice to other games after learning the rules and limitations?
better yet can we transfer its ability to humans?
Essentially it can. It learned chess in 3.5 hours and beat the previous best chess bot (Stockfish) without losing a single game.
Is AlphaGo made in Go (golang) - Google's programming language?
No yo
"you scared, Nash" "mortified, petrified, stupifided, by you"
such a bold statement saying theres more configurations than atoms in universe. same board with one more row already have more conf.
This makes absolutely no sense. No-one claimed that there were no games that had more configurations, only that the number of configurations was more than about 10**80.
Go is truly an amazing game. (:
Lee Saedol is an interesting choice: he is the most unpredictable and creative player at that level. Commentators have hard time "explaining" his moves.
The Machine from Person of Interest is coming (;
They'll be beating us at Naughts and Crosses next.
SAI IS BACK! He's trapped in a computer!
Nathan Rogers Here I am a year later desperately looking for a comment about Sai!!! and i finally found it lol
But can it win on a cold rainy night at stoke?
So what happens when AlphaGo plays itself? What can we learn from such a match?
the same thing it learns from other matches, not strategies that can increase the win percentage.
the first general artificial intelligence would come out of puzzle games that mimics human speech and problem solving. If we make a game complex enough with the right human reinforcement learning; a game that f.ex. requires to do hard from scratch coding or solving a physics problem. Right now we don't have an architecture for this but soon enough we will lay the ground works for the deep-mind to improve to become Artificial general intelligence. The key is making games that are challenging enough such that it reflects some part of human abilities. Hold on tight it is going to be wild in the next decade or so.
Ah this game
If it can beat a professional go player it can dodge a wrench. XD
Fan Huei looks pissed!
Nice to know how large a handicap the human world champion will need to get an even chance to the computer champ.
I have a theory that since I have no previous experience with go and have no idea at how to go about it, I wonder the outcome due to the fact that these think on the quantum level
Aka a complete different set of rules of relativity and these devices think on our theory of relativity I ponder on the fact that since a player with no experience may do better compared to the greatest go player based on that logic I would love to conduct an experiment on that basis if you would be interested please dm me is love to find these answers
Due to the fact that AI is artificial intelligence,it is indeed what it says intelligence based upon previously gained knowledge compared to agi (artificial general intelligence) intelligence fed In to a machine and is able to find the best absolute/logical outcome
On its own
History in the making
alpha go beat lee sedol handily
"More configurations in the board than atoms in the universe" - this is just a phrase to impress. A string with 60 latin characters has more configurations than atoms in the universe. Does this make a text string particularly interesting o difficult to treat?
4:15 finger cut
Wow. We are handing our world to the computers just like that. It's crazy!
Who else just watched/played the warframe sacrifice quest
Isn't CrazyStone A.I. a brute force approach to Go?
*☼ the same way we play chess: stupid algo cookbook bs on best guess odds.*
input status of board, output next move.
New era
A European champion is no Lee Sedol or Go Seigen. I'll be interested to see the match against Lee Sedol.
this software would be a good foundation for skynet
This is chess on steroids.
Next machines will make great jokes.
And soon enough they will excel at everything we do.
And we will turn into them as we touch the stars.
mastertheillusion you should take a look at the channel "Nicholson 1968". He knows a lot about transhumanism, the same thing you are talking about. But he's not so optimistic.
but that AI can predict something like human means it could think like a human
if you know what is this game you would understand
like millions of changes could happen in this game
But can it dodge?
4:05 A computer program that can beat high level go players is a very good achievement, the developers should be proud. But in what way is it akin to imagination? It's an algorithm, it computes numbers, it is only useful in a particular sets of conditions. Imagination is orders of magnitude more general.
Of course, nobody taught them what a personality (or soul) is. What is it btw? AlphaGo has never played Chess, so how good can it be at it. But... the *algorithm* that learned to play Go extremely well, would most likely learn to play Chess extremely well, or maybe with some more tweaks and improvements learn to drive a car through traffic, or guess the best stocks to buy or sell or etc. etc.... The "imagination" in us humans is in fact our own *algorithm*, which is implemented in our *hardware* (the brain) just like AlphaGo's DNN was implemented in some computer cluster.
When will we see the first commercial go bot to beat pro players?
They must make a app with that I'll buy it
He is beating Se-Dol Lee(Former champion of this game) with score of 2-0 out of 5 games series. The most scariest thing is that commentating pro players couldn't even follow the move computer is doing, thinking its moves were often faulty. Only the player himself seemed shocked by creative and shockingly effective unconventional moves. gall.dcinside.com/board/view/?id=baduk&no=64184&page=1&search_pos=&s_type=search_all&s_keyword=%ED%95%9C%EC%88%98
When play yu gi oh ?
If it makes anyone feel any better. A.I. still has a long ways to go in video game play. An elite level Donkey Kong or Moon Patrol player would still easily beat AphaGo.
they've got us beat in Tetris though.
I enjoy
Mucho gusto
@@josephthibault798 gaspacho Alvalle
@@georgesguitton8358 La soupe froide qui vient d'un pays chaud