AI vs. Human: The Greatest Go Tournament Ever
Переглядів 631,630
8 років томуGoogle's 'AlphaGo' and the world's top ranked Go player go head-to-head in a battle to decide whether or not an AI can be programmed to win a game as complicated as Go.
Link to match videos: / @google_deepmind
Hosted by: Hank Green
----------
Support SciShow by becoming a patron on Patreon: / scishow
----------
Dooblydoo thanks go to the following Patreon supporters -- we couldn't make SciShow without them! Shout out to Justin Ove, Accalia Elementia, Kathy & Tim Philip, Kevin Bealer, Justin Lentz, Fatima Iqbal, Linnea Boyev, Tomasz Jonarski, Chris Peters, Philippe von Bergen, Will and Sonja Marple, and Mark Terrio-Cameron.
----------
Like SciShow? Want to help support us, and also get things to put on your walls, cover your torso and hold your liquids? Check out our awesome products over at DFTBA Records: dftba.com/scishow
----------
Looking for SciShow elsewhere on the internet?
Facebook: / scishow
Twitter: / scishow
Tumblr: / scishow
Instagram: / thescishow
----------
Sources:
www.nature.com/nature/journal/...
googleresearch.blogspot.com/20...
www.wired.com/2016/03/googles-...
www.wired.com/2016/03/go-grand...
www.theverge.com/2016/3/13/111...
qz.com/636637/the-beginning-of...
www.techinsider.io/why-google-...
Image Sources:
commons.wikimedia.org/wiki/Fi...
commons.wikimedia.org/wiki/Fi...
commons.wikimedia.org/wiki/Fi...
@kunwoododd2154 4 роки тому
The Go experts invited to watch the fifth game as it played out had commented on how at first they were confused why AlphaGo was making so many rookie mistakes, but later on they realized that AlphaGo was deliberately aiming to win the game by only 1.5 points. They then confessed that they had just realized they were witnessing an AI play Go at a level higher than their own comprehension.
@GrahamCrannell Рік тому
that's a pretty key difference between AlphaGo and human players. AlphaGo only wants to win, not necessarily win by a wide margin. It will accept a half-point win as being the "correct" outcome, whereas a human player would strive to widen the margin
@eragon78 Рік тому
@@GrahamCrannell Especially in Go, since it can be extremely difficult to tell who is winning at any given point, and battles you thought you won, or spots you have a large amount of points in, can easily change based on how the game is progressing. So Humans always want a big buffer with lots of points to maximize their odds of winning even if one or two of their battles they thought they were going to win they end up losing. A computer like AlphaGo though has much better prediction powers since it can calculate way more moves, so it can more accurately assess how each battle may go meaning it can more accurately plan its margin of victory by going for less risky plays and just guaranteeing it gets enough points from every battle to just barely win.
@crabapple. 8 місяців тому
Humans.. are bound to greed
@katzen3314 8 років тому
I think it's more impressive that somebody could beat it...
@ching9679 8 років тому
+Katzen4u His dad probably beat him after the match because he only won once against the most developed AI in the world :(
@katzen3314 8 років тому
Kevin Pranoto I hope not lol.
@MrWizardjr9 7 років тому
no
@ThePROpcgaming 7 років тому
I doubt that this is the most complex engine, there some chess engines that are out of this world...
@katzen3314 7 років тому
I thought the best chess AI ever built was a neural network. They can have really impressive results. But yeah, they can be relatively simple.
@TheArzonite 8 років тому
Whenever you play chess against computer and win, it's because the computer lets you win. Feels bad man...
@Teth47 8 років тому
+TheArzonite Yeah, it's pretty mindblowing that we managed to make a thing that performs a mental task we invented for ourselves better than we can...
@thelaw8042 8 років тому
+Teth47 These tasks weren't made for us. We could play them is all. We MADE these programs for the game.
@miguelrealp 8 років тому
+Teth47 Why is it mindblowing? They are doing what they were created to do. A wheel spins, that's why we created it for, It's not mindblowing at all
@conduit64 8 років тому
+TheArzonite Actually they still haven't made an AI that can win at Chess 100% of the time and programmers have been working on Chess AI for decades.
@shamtradtam3769 2 роки тому
@@thelaw8042 then who the hell is chess made for? Aliens?
@kurtilein3 8 років тому
One interesting fact about the strength of Lee Sedol: Near the end of the 5th game, Lee Sedol calculated every possible continuation all the way to the end of the game, counted the territory of both sides, and figured out that he was 1.5 points behind, and resigned. 1.5 points on that gigantic board is a tiny difference, also Lee Sedol was out of time and only got one minute per move, so he did the calculation in less than a minute. So the elite go professionals are insanely strong. Against weaker players they can intentionally play bad moves, and if the weaker player continues to try hard, they can force a win by exactly 0.5 points. They can also play blindfolded. They can also play multiple simultaneous games against several opponents.
@iAteYourDog69 8 років тому
uhm, k?
@RanaKamikaze 8 років тому
+kurtilein3 Asians...
@speedy01247 8 років тому
+kurtilein3 so it is like chess, but harder just with fewer different pieces.
@Joeyyukonm 8 років тому
Same with some chess grandmasters.
@kurtilein3 8 років тому
speedy01247 the rules are easier, its easier to learn the basics, but with 361 points where you can place a stone, the emergent complexity is much higher. Memorizing the entire game state or an entire game is also much harder. Google had to use a supercomputer, and the program is much more complex than a chess engine. Different parts of the program do very different things and work together like a team. For example, the program needs a massive database and a massive neural net, without these it would be weak. For the near future, it seems impossible to get a strong go program to run on something like a consumer-grade workstation, even if it is a powerful workstation.
@lazyhusky502 4 роки тому
At 3:08, it’s actually that he saw a move that AlphaGo didn’t read into too deeply because it’s a move that 1 in millions of Go players would actually see and be able to read out. Therefore, it didn’t see that move as a threat. The only reason I’m commenting this is because it really shows just how good of a player Lee Sedol was, particularly at fighting.
@drizzify7850 2 роки тому
the documentary said 1/10,000 if I remember correctly
@DanielRenardAnimation 8 років тому
Somehow, the part of the story about the Wedge move made me _happy._ Showing that we humans still have tricks up our sleeves, when faced against something artificial, designed solely to "destroy" us.
@uegvdczuVF 8 років тому
+Daniel Renard YOLO moves work in games much better than in real life. :)
@mimsydreams 8 років тому
+Daniel Renard The move we have is the fact that we build and program these machines. Just smash it with a hammer if you think it's going to kill you by winning at a board game o.o
@rdizzy1 8 років тому
+mimsydreams Until it is armored and has missle launchers and other firearms, haha.
@rdizzy1 8 років тому
+mimsydreams Until it is armored and has missle launchers and other firearms, haha.
@mimsydreams 8 років тому
rdizzy1 We probably shouldn't give the board game playing robot armor, firearms, or program it to use such things. I don't see that as much of an advantage at playing board games.
@BattousaiHBr 8 років тому
I, for one, welcome our new computer overlords.
@bobbiusshadow6985 3 роки тому
"All hail our AI robot overlords"
@GrahamCrannell Рік тому
I was a graduate student doing research in machine learning at the time when AlphaGo beat Lee Sedol. My advisor and our research team basically got *nothing* done for about a week while we discussed and analyzed the games and the deep-learning algorithms behind it. Fun times 😄
@darkwoodmovies 8 років тому
I read the paper they published. DeepMind (the division of Google that created AlphaGo) has made some amazing innovations in deep learning.
@pramitbanerjee 8 років тому
+darkwoodmovies link?
@darkwoodmovies 8 років тому
willamette.edu/~levenick/cs448/goNature.pdf
@pramitbanerjee 8 років тому
thank you, i should have realized this would be beyond my knowledge comfort zone right now. But one day i will learn computational neurobiology and then i will read this paper.
@MartinThoma 8 років тому
+pramitbanerjee It's not "computational neurobiology" but machine learning with neural networks
@pramitbanerjee 8 років тому
Martin Thoma can you explain it better, what is a neural network and how is it different from the neural microcircuits in our body? I understand machine learning involving computational algorithms and that knowledge is beyond me atm
@markzeddo6033 8 років тому
The wedge wasn't to confuse Alphago. It was what we call a tesuji: a very clever move that you might not think of right away. And the interesting thing about that wedge is that it made the capture of some other stones in the center possible, which meant that Sedol had a few more points than he would have had, and Alphago had almost none in the center. Michael Redmond has a good explanation as to precisely what that wedge does. But the goal of that move was to capture stones, not to confuse the computer.
@ShirinRose 8 років тому
Thank you for reducing the volume on the video suggestion preview at the end! Scared the living daylights out of me before XD
@drakan4769 8 років тому
3:28 well it still won 4 out of 5 games agains the best human around, so it might not be perfect, but right now I'd still say it earns the "best" title
@gamarad 8 років тому
+Drakan R Lee Sedol is currently ranked number 4 in the world. The number one ranked player, Ke Jie has a record of 8-2 against Lee Sedol so he has the same win rate and he thinks he may have been able to do better. I do see your point though and the version of AlphaGo that played Lee is already out of date so it's a moot point.
@litojonny 8 років тому
interesting! BUT WHY DO I HAVE HAIR AROUND MY ANUSSS??????
@christianharrs9268 8 років тому
It means you have hair follicles in that general region. Or was that just a troll post?
@Por-poI 8 років тому
+Christian Harrs you don't hang around in scishow videos comment sections often, no?
@Por-poI 8 років тому
+JpegXguy it's more of an easter egg than a joke, actually. when people say that others would instantly know that they watch Scishow videos a lot. much like "Hey I'm a scishow fan" "Me too!" *fistbump*
@Por-poI 8 років тому
Free cat for good home yeah, exactly like that. haha
@SheezyBites 8 років тому
As a GO player it was really interesting to watch these games. Both AlphaGo and Sedol made mistakes or weak plays at times (that first game had a big mistake for Sedol), but it was still amazing to watch such tight games. I'm excited to see where AlphaGo can take the game and see what strategies it can come up with that human players haven't.
@jeff86ing 8 років тому
This year AI masters Go, next year the stock market.
@ahtoshkaa 8 років тому
+Peanutbudda this actually sounds very real to me. With the google's computer power and resources they might actually try that.
@GekoPoiss 8 років тому
+Peanutbudda Already is
@Roxor128 8 років тому
+Peanutbudda There are trading bots for stuff like foreign exchange trading right now. Heck, some trading platforms even let you write your own.
@saintebetterave3001 6 років тому
Today trading bot only invest in the short term,
@lizciz 8 років тому
I just had my exam in a course in Applied Artificial Intelligence two days ago, so this video was really relevant and interesting for me! Thanks!
@ChessNetwork 8 років тому
Yet another game schooled by an AI. :D
@jojo300001 8 років тому
+ChessNetwork HI JERRY~
@JediBladeMaster 8 років тому
+ChessNetwork Hey! It's you. :D
@DrewKF 8 років тому
Hai! I don't know who you are but I'm now under the impression I'd be terribly excited to see you posting if I did know who you are and had a very particular set of interests akin to those of the above posters. As I do not, this is all just mildly exciting... woo! :D
@qwertz12345654321 8 років тому
+ChessNetwork I wonder how a Dead mind chess engine would do against Stockfish and similar...
@GreerFried 8 років тому
+ChessNetwork OMG Hi!
@Spironic 4 роки тому
3:28 "So AlphaGo might not be the best Go player in the world right now" This is a really strange choice of words considering it won 4 out of 5 games against the best human player in the world. It would be more accurate to say it isn't infallible and doesn't play perfectly/isn't unbeatable, but it's still the best player in the world.
@Marcsine 8 років тому
So proud of AlphaGo! Ever since I learnt AI on UKposts, Go was *always* to goto example to show how computers generally can't solve complicated modern day problems.
@Beykakua 8 років тому
THANK YOU FOR FINALLY TALKING ABOUT THIS!
@danielnorton9803 8 років тому
This could make a good modern episode of Hikaru no Go!!
@MariE-bz2eq 7 років тому
Daniel Norton I was thinking the same thing. I remember in one of the later episode, the chinese guy who was helping isumi stating that the hand of god move would be found through a.i.
@AX5Terminator 8 років тому
Google needs to lend a helping hand in the video game industry, because the AI in video games are still a long way off from being perfect.
@greoge1381 8 років тому
+General Ivan Total war...... JUST FUCKING PUT THE LADDER ON THE WALL!!!!
@Poctyk 8 років тому
+General Ivan Would you want to play against AI that after 20 or so games would beat you every time?
@AX5Terminator 8 років тому
+Ростислав Несисюк well they could always scale them back. the point is to be able to make the AI appear smarter.
@StevenDavidson 8 років тому
+General Ivan I'm not sure what you mean here, hell the AI of some video games 5 or 10 years ago was pretty good in many cases. Besides, do you really want 4 titans running constantly just to have AI followers that won't occasionally run into a wall?
@Saposhiente 8 років тому
+General Ivan Problem is that the AI's objective in most games isn't to be the best possible player--too punishing of an opponent and too dominant of an ally--but instead to make sensible decisions, creating difficult but winnable situations as an enemy and cooperating well as an ally. The AI training is based on maximizing score, learning which moves will increase your final score, but these objectives are much harder to measure in a score.
@violin2971 8 років тому
Thank you Hank! I was trying to follow this story on Wired but I don't know how to play go and was generally very confused.
@FernandoPereira88 8 років тому
I love the fact that this amazing channel has English subtitles. I would like to join the staff and add Portuguese subtitles to this, but my English isn't that good yet.
@frankschneider6156 8 років тому
+Fernando Pereira Doesn't seem to be too bad. So, why don't you try ? Even faulty ones, might be better than none.
@prodigypenn741 8 років тому
hikaru no go was right . . . the path to the divine go lays in computer technology
@rumitoowavy8711 4 роки тому
prodigypenn thought I was the only one here who seen it
@megamaniscoolrightguys2749 4 роки тому
Woah...😳 I never thought of it like that!
@lancelindlelee7256 8 років тому
If it keeps learning though, it will eventually be unbeatable.
@HummingbirdCyborg 8 років тому
Glad to see a Scishow video about AlphaGo defeating Lee Sedol. It's a sad day in one sense for those of us who love the game, but in another, it's exciting because you can imagine a similar system improving our ability to perform other tasks that have been hard for AI to master.
@computerchi 8 років тому
never before Go had so much media coverage. let's look at the bright side.
@frollard 8 років тому
I love how in a game that complex one 'wedge' move can be seen as a turning point where it all went bad.
@Roenazarrek 8 років тому
If alphago wins 4 out of 5 games with the best human player in the world how is it not the best Go player in the world?
@jivjov 8 років тому
Because Sedol's record is still better than AlphaGo's
@rdizzy1 8 років тому
+jivjov So? In every other sport or game one can think of, record doesn't matter much, if you beat the best, you become the best. I mean they could've easily fed lower ranked players to GO hundreds at a time to pad up its record, which wouldn't make it any better.
@jivjov 8 років тому
+rdizzy1 you're the best for a particular year or season; not "best of all time"
@jivjov 8 років тому
+UpToYouOnly and you're tossing around ableist slurs......why?
@rdizzy1 8 років тому
jivjov Nah, if the person you beat, consistently 3 times in row, has already beat the best of the best over the past 10 years straight, you become the best of all time. (Noone obviously ever means that literally, as noone could ever prove they were the best in a thousand years through billions of individuals.)
@Andytlp 8 років тому
hair in your nether regions is to soak up sweat and reduce friction.
@suntzu8642 8 років тому
Thank you . I just got enlightened😇
@inferno7181 8 років тому
Still don't understand. I need SciShow to explain.
@deathpony698 8 років тому
+tulp35000 way to play along
@IvaNiftyChannel 8 років тому
+tulp35000 you didn't even blend in letters on a green screen, how am i supposed to understand!?!?
@Andytlp 8 років тому
yup Hank, cover this more in depth. pheromones and all.
@edmundthefox3656 5 років тому
Thanks to learning about news like this, Go is now my absolute favorite board game.
@BarakaGitari 8 років тому
Thats amazing. That these were close games, man vs machine and all
@1006christopher 8 років тому
Now I just have to learn how to play Go to see how good it is.
@idioume1 8 років тому
+Christopher Setterstrom IT's awesome
@SpiteBellow 8 років тому
Go seems cool. Never played but i want to try it now!
@SmileAdina 8 років тому
+SpiteBellow you should! ;) Search for a go club in your town. You'll probably find one.
@computerchi 8 років тому
thanks. this was a very good summary.
@OperatorZx 8 років тому
Just for the record, Lee Sedol was only the top ranking person from 2007-2011. The current top player is Ke Jie who beat Lee Sedol 8 out of the 10 times they have played.
@jlw9113 8 років тому
I'm a CS major and a Go player of a few years. Maybe people will pick up this awesome game.
@MrBlitz2000 8 років тому
+Atticus Weiss I'm trying, but it's so difficult. :(
@kilo1012 8 років тому
should have had the ghost of Sai fight the AI
@FakhriAhadi 8 років тому
But Hikaru drove him away
@owltalktoyoulater3940 8 років тому
+Fakhri Ahadi don't remind me ;-;
@aacmack 8 років тому
Someone write a fanfic!
@owltalktoyoulater3940 8 років тому
YES, well I've got my next oneshot set out for me then
@mikiekwoods 8 років тому
Great video, although lee sedols brilliant move in game 4 wasnt because it confused the computer but because it was just really good.
@randomnoob6524 5 років тому
At first I thought it going to be the game play of it since I play go and I love it.
@ReeCocho 8 років тому
SciShow: DNews, except not garbage.
@LinJukin 8 років тому
How is Dnews garbage?
@ReeCocho 8 років тому
Because they talk about stupid shit like "The War on Pubic Hair" and "Should Guys Freak Out About Their Penis Size?" just to name a couple.
@yassinefujiwara8247 8 років тому
+Lin Jukin lacy green
@ReeCocho 8 років тому
Yassine Tama Not going to lie, I was going to put that at first but then I said to myself (But then I'll be part of the patriarchy! XD)
@InorganicVegan 8 років тому
+Lin Jukin Also, watch their video on gross food ingredients. Basically, the logic is "Hey, carrots have hydrogen atoms, so do H-bombs. EWWW!"
@ziquaftynny9285 8 років тому
This is it, I am living in the age in which A.I gets big! Fuck yeah!
@empresslithia 7 років тому
More amazing is living in the age that genetic engineering gets big.
@joaodecarvalho7012 7 років тому
It is hard to predict which of them will be bigger. So far AI is much more present. CRISPR promises to heat things up in the genetic revolution, but its products are still to have impact in everyday life.
@empresslithia 7 років тому
Rather than genetic engineering, we might also see a boom in cybernetic augmentation (robot limbs) but that's not as likely I think.
@40paschal 7 років тому
just like everything else we need to let productions of scale enable the rest of the population to afford something like CRISPR.
@mueezadam8438 8 років тому
You should include some of the initial statements from the top players at the time of Sedols' loss, they're quite interesting. Lee Sedol himself saying that _"this his my loss, not humanities"_
@curerz 8 років тому
The AI is just like Komugi, always evolving after every match to learn more. 10/10.
@ModelOmegaForReal 8 років тому
I for one welcome our new computer overlords.
@TheHeadHunter105 8 років тому
+ModelOmega Spotted the android sleeper agent
@TheVariableConstant 7 років тому
The fact that it lost game 4 is incredibly important and nullifies the fact it won the first 3 games. Basically, alphaGo plays against itself multiple times plus it beat the EU champion 4 out of 4, before winning first 3 against Lee Sondol. At this point the A.I. has far better "inventory" and it should never lose again, yet it did. Also the human brain is doing far more things outside playing Go, like regulating heart beat, maintaining posture, coordinating movement, chemical homeostasis, listening to the environment etc.. than it is doing playing Go. Yet it was able to win even 1 match. It shows just how far advanced human potential is. Even during the Go match the mind only dedicates a fraction of computing to winning so imagine life away from the Go board? Meanwhile AlphaGo is always and only working on improving it's Go skills.
@theletterwynn 7 років тому
The human equivalent would be having a surgically grafted robotic system of internal organs and a brain upgrade while chefs and nurses fed and cleaned you so all you needed to do in life was play Go. Pretty fun life - maybe.
@MrWizardjr9 7 років тому
what how does that make sense. so if lee sedol won every game instead of one that loss nullifies the other wins.
@Aquaified 7 років тому
TheVariableConstant clearly you have no fucking idea how machine learning works, or how anything works because you must be brain dead. The few matches AlphaGo won against the top players didn't add much of anything to its algorithm. It trained on thousands and thousands of games. You really think those few matches give it some incredible "inventory"? Jesus Christ that makes no sense you're a moron.
@grandpahiggins 7 років тому
Aquaified well when chess computers were being made for commercial competition they played against top players to debug it's system so I don't know how important it is to the system but it can be helpful I don't know if I agree with his comment of nullifying the other games the computer won but human competition is good for new and upcoming programs just to give some variety send tho later on we prove to be a significantly less in strength than a computer in terms calculation our consciousness can provide a decent fight and at times virtually equal
@Aquaified 7 років тому
grandpahiggins of course real matches were used, that's not what's wrong. What's wrong is thinking that just a few matches would make any difference whatsoever. Those computers were trained on thousands and thousands of matches.
@scoopscornchip74 8 років тому
this is amazing
@cyestoner 8 років тому
Move 78 in game 4 was not intended to confuse AlphaGo! It was legitimately the best move available to Lee Sedol. NOBODY else (including other top professionals) saw that move or considered it, but it absolutely worked without resorting to any sort of "confusion". The fact that AlphaGo chose to play really stupid moves after that is more of a problem with Monte Carlo methods than an indication that AlphaGo was "confused" by the move.
@idkrossplay 4 місяці тому
It didn't really play stupid moves, it calculated all of the forcing lines in the center that lose so it tried to ignore it and expand elsewhere. AlphaGo calculates its probability of winning so it saw that trying to stop white would be a guaranteed loss
@mikeunleashed1 8 років тому
BUT THE CHAMPION WON AT LEAST ONCE! THAT IS AWESOME, YES CAPITALS ARE REQUIRED!
@Myrslokstok 5 років тому
mikeunleashed1 Maybee it was to smart to outshine the master.
@888SpinR 8 років тому
Is there a chance that perhaps the reason humans keep losing to the computer is because the computer is designed to play against humans, but the players weren't "trained" to play against AI? That fourth game seems to suggest that you just need the right strategy for playing against computers. While we're at it, how would a game play out if two separate AlphaGoes were to play a match against each other?
@Plystire 8 років тому
+888SpinR Didn't the Ai teach itself based on games it watched, and then played against itself to improve? Technically it was only ever trained to beat itself, and it just so happens that it trained to be better than humans because it was heuristically better at the game than humans. Though I'm sure you can find a pattern in its play style and eventually exploit it for a bit before it gets even better. :P I mean, this thing was designed to think *long and hard* about what it was doing to win at this game.
@mueezadam8438 8 років тому
Little correction hank: it was the first time a computer beat a professional player without a handicap (placing a set amount of stones of the weaker player's colour before the game is played) on the 19x19 board. on the smaller 9x9 boards however computers have been shown to beat pro players occasionally.
@ayreseatsdubia 8 років тому
I have been playing since I was 4, but I ended up losing interest after I stopped playing for a while. Then I heard about alpha-go and the Google deepmind challenge. I watched all the matches, and they were amazing; all the moves were so well thought out! It got me playing Go again! :)
@MetaBloxer 8 років тому
+Jessica Leung AI=An Inspiration
@ayreseatsdubia 8 років тому
+MetaBloxer ikr
@elimartin3774 7 років тому
Jessica Leung, I love go. There's a simplicity to it that makes it more entertaining than Chess.
@ayreseatsdubia 7 років тому
Eli Martin it's simple, yet complicated in a sense. amazing game.
@iwannacomein 8 років тому
But can A.I. beat Komugi at Gungi? The answer is "No".
@chainezo 8 років тому
+iwannacomein D: the feels man
@jheffstrongman6633 8 років тому
+iwannacomein i love you who ever you are i'm not alone
@ivyy11 8 років тому
LOL i thought i'm the only one thinks that
@john3260 3 роки тому
When you put your life at stake to guarantee yourself winning every game...
@DrHotelMario 8 років тому
Great... now make it play global thermonuclear war
@JupitersDancer 8 років тому
+DrHotelMario It's bad to play wargames :P
@aidannagle1488 8 років тому
+DrHotelMario The...only...winning...move...is...not...to...play...at...all.
@Jackboy019 8 років тому
+Aidan Nagle Well to be fair, since the AI was able to win in a game with a huge amount of possible moves in Go, it might be a real possibility for real life AI to simulate the game against itself to find the best method of winning in a global thermonuclear war.
@bobbiusshadow6985 3 роки тому
let's play tic tac toe first
@rbpompeu1 3 роки тому
Hahahahahahhaha
@jackpeterson6295 8 років тому
I hope this will give go more popularity in America. Whenever I mention that I play go to someone they don't even know what I'm talking about and I have to explain it to them. Go tournaments are massively popular in Asia, but the US Go Congress only gets hundreds of people each year.
@1997jankuschef 8 років тому
Awesome. Just awesome.
@zeromailss 8 років тому
Alphago make a mistake,that actually make it even more like human
@revampedharpy09 8 років тому
yeah... and thats something i dont like. dont get me wrong, im happy we CAN do it, but us actually doing it, not so much. ever watched the terminator movies. well a large amount of experts in AI are saying we could be dealing with a REAL skynet as soon as 2045, only 10 years after NASA is putting people on mars in 2035.
@zeromailss 8 років тому
revampedharpy09 at this rate,yes it should be around that time,but look,even from few years ago a lot of scientist already discussing on how would an AI behave if they ever get self consciousness and how to prevent them from harming human,with 30 years we should be able to come up with some decent answer or humanity is doomed or not,its still unclear whether a machine could get self consciousness anyway and above all we human still not 100% sure what define consciousness let alone make an AI with one
@revampedharpy09 8 років тому
Yukino Takada 雪乃 鷹だ personally i think we should build EVERYTHING with ANY AI with a remote kill-switch incase it turns evil so we can just shut it off immediately
@zeromailss 8 років тому
revampedharpy09 that is one of the possible answer,but if an AI that is as smart or smarter than a human then it would see the off button/program as a thread an it might even hack its own system , you never know
@revampedharpy09 8 років тому
Yukino Takada 雪乃 鷹だ thats true but surly thats a risk worth taking, i mean whats the worst thatll happen? itll remiove the kill switch and we get left with the same situation as if we didnt install the kill switch at all so we dont get anything worse than we would have anyway.
@oscarr0_794 8 років тому
The future of sports... Coding the best ai to play for them
@rdizzy1 8 років тому
+Oscar OR just coding the best possible plays against certain opponents, could work in many sports like football, or UFC fighting.
@sabahancover7566 2 роки тому
I'm developing a very great interest in this board game because of Hikaru No Go😍 I'm loving it!!
@PhilippeCarphin 2 роки тому
The words "deep learning" or "neural network" could have been thrown in. It's somewhat the key factor in surpassing humans in go. And the fact that there are two neural networks, the policy network and the value network would also have been interesting to mention.
@Censtudios 8 років тому
I'm telling you. Google will become skynet. You can quote me in a few decades
@spencergeller2236 8 років тому
+jokicha HE'S FROM THE FUTURE
@reallygoodbook7258 8 років тому
+Spencer Geller From a few decades in the future, to be more precise
@muhammadaizatcheazemi9062 8 років тому
So in the future did Holland ever win the world cup? Or Trump become president? Or Leonardo DiCaprio win an Oscar? Oh wait...
@robertbones326 8 років тому
+jokicha "+Censtudios "I'm telling you. Google will become skynet. You can quote me in a few decades" -Censtudios" - Jokicha
@DrewKF 8 років тому
+Muhammad Aizat Che Azemi Trump wins an Oscar and DiCaprio... well, he would have been a great president if it weren't for the assassination, hours before his inauguration. I may have said too much. Back to the Delorian!
@WrellMM 8 років тому
here comes Skynet
@adamwestpc 8 років тому
dope, never thought that deciding between a good move and a bad move could be an explicit state
@thatchessguy7072 2 роки тому
@3:08 The reason why the wedge worked wasn’t because there were too many response to the tesuji to calculate how to save face. The sequence of moves that were required to escape the wedge required that the computer make bad shape. This went against the grain of the system (it was not reading out bad shape moves) and it went into a Monte Carlo crash for a bit. Newer go AI are able to cope with this wedge. The reason why the position was complicated was Alphago lee shoulder hit a 3rd line stone which heated up the right side and caused the fight over the center.
@Kombaiyashii 8 років тому
"And spoiler alert, Alpha go won in three." That's no spoiler alert. Give people a bit of time.
@akeemperez8509 8 років тому
can't wait for a computer to take my job, I worked so hard on my degree so it's so thrilling to know a robot will be able to do my job in half the time and get paid nothing to do it......I love computers 😭😭😭
@ordinarytree4678 8 років тому
+Akeem Perez The next step in human evolution is robotic augmentation!
@Teth47 8 років тому
+Akeem Perez So move up the ladder. Computers will take the basic jobs first. Now you must work smart, not hard.
@akeemperez8509 8 років тому
Teth47 what about everone else not everyone can be at the top. Do I look down on them for not achieving what I have accomplished? Do I say do what I did for the position I have even if only one person can hold the high seat at a time?
@ordinarytree4678 8 років тому
Akeem Perez The poor will keep getting poorer until another revolution happens. Thats how these advances always go.
@Teth47 8 років тому
Akeem Perez The more robots fill in the lower positions, the more high up positions there will be. Robots don't cost as much, freeing up money to be invested elsewhere. The amount of money in the system continues to grow, meaning that even if the proportions stay the same, everyone gets more.
@herimusiclass 5 років тому
I can just imagine the frustration the champion is feeling, he's probably he's practicing like crazy
@arousedbanana6327 4 роки тому
Go has like a number of moves that goes beyond atoms.
@erictu4900 8 років тому
can they beat good players at cs: GO ?
@Teth47 8 років тому
+Eric Tu Depends what we let the computer see, but almost certainly easily. CS:GO is more reaction time than anything, the AI would just sprint and spin around as fast as the game will let it, the moment is sees a group of pixels that's likely to be a player, it puts a bullet or 40 there.
@verdiss7487 8 років тому
+Teth47 yeah not even a contest
@stevenboelke6661 8 років тому
+Teth47 Aim bot is a thing so I imagine they just need to make it walk from a safe position to another.
@Pichipieify 8 років тому
well, im GE but i still cant beat a spinbot
@Darkstar_64 8 років тому
I think that would be the greatest vac case ever
@DarkBioCloud 8 років тому
whats it going to take for humans to realize it is not a good idea to teach computers how to beat us
@BottleWaterson 8 років тому
when we program something to beat us at something that isnt a board game.
@kunphushion 8 років тому
+DarkBioCloud There's a lot of energy and resources going into making robots not take over the world.
@pakwakalexina2345 8 років тому
.
@CeoLogJM 8 років тому
That's easy, we just make something else that would kill them!
@katzen3314 8 років тому
I bet in the future we'll all live in fear of the evil AI controlled robots that can play board games really well...
@user-je7db3hn6m 8 років тому
Think AlphaGo was shown THOUSANDS of human-played games, then set to play against itself 30,000,000 times, nor the other way around. In any case, many, many thanks for all the videos. I am a fan.
@Kaoskadosk 8 років тому
What's really interesting is that Lee Sedol said in an interview that AlphaGo wasn't a better player, but it never got tired. Since it's a computer it can use it's full capacity for every move while a human becomes fatigued, so as the match went on, AlphaGo bit by bit gained the upper hand by "wearing down" Lee Sedol. Why this is so interesting is that AlphaGo wasn't created just to see if an AI can beat a professional at Go, but to use the knowledge gained from that achivement and apply it to e.g. medicine. If a machine can perform complex tasks like Go at the same level as a human without becoming fatigued, that's a major thing. If an AI like AlphaGo could replace doctors maybe there would be fewer misdiagnoses due to human fatigue, which currently is a rather large problem.
@DatHamTho 8 років тому
WHY ARE THERE ANUSES AROUND MY HAIR??????
@General12th 8 років тому
+DatHam Because you're a butthead.
@EnigmacTheFirst 8 років тому
I swear, if there isn't an April fools day video about why there's hair around my anus.
@suntzu8642 8 років тому
Why do I see it everywhere on that comment section ? 😂
@GammaProtogolin 8 років тому
+Αργυρης Χατζηηλιας I know right?
@suntzu8642 8 років тому
+Doomsdaywillcome is it like a troll😝
@inferno7181 8 років тому
+Αργυρης Χατζηηλιας fuck off. Emojis are cancer.
@zaynevon-kai6090 8 років тому
+inferno As is Pokemon.
@TheJpmuzz 8 років тому
at the 0:26 mark when Hank says "When engineers talk about AI, they don't mean sentient humanoid robots" he is dead wrong. He is only right about the robot part. AI is by definition "The field was founded on the claim that a central property of humans, human intelligence - 'can be so precisely described that a machine can be made to simulate it.'" & "of creating artificial beings endowed with human-like intelligence"
@thegoombrat8198 8 років тому
So in short in these past few weeks we have taught AI how to out think us whilst at the same time bullying what are basically early terminator. Well nice knowing you humanity.
@fartonaut2291 8 років тому
Google has a thing with programs that "learn"
@adtc 8 років тому
Isn't crawling the web so that you can "Google it" a form of learning?
@Stray0 8 років тому
Next time you say "Spoiler Alert", could you wait 5 seconds so I'm able to react to that?
@biliboy89 8 років тому
Is it possible a video about the wow noise? never understood why it is wow!
@lake4ishikawa 8 років тому
I just have to say this since I hear this "there are more Go positions than there are atoms in the universe" riddle everytime: 10^170 is not just "more" than the atoms in the universe (ab. 10^80), you could replace every single atom in the universe with another entire universe, and there still wouldn't be enough atoms to cover all possible board positions, in fact they would still be 10 billion times less.
@natsudragneel7342 8 років тому
I bet it can't beat kimogi.
@SuperSox97 8 років тому
+Natsu Dragneel It's spelled Komugi but it was obvious who you were talking about.
@reorx9 8 років тому
+Natsu Dragneel definitely not since the go program would have no idea how to play gungi, however if you thought it was in a go game then meruem would be a better candidate since we know meruem has skill in go while we dont know about komugi
@dentescare 8 років тому
If that AI could play gungi, it would be great. it's impossible anyway but whatever.
@kunphushion 8 років тому
+dentescare nothings impossible for A.I. eventually.
@Desmaad 8 років тому
Go doesn't have have hugely complex rules, but the theory for it is well-deep.
@Jackboy019 8 років тому
I think AlphaGo's weakness when the wedge came up could have been solved if the AI knew how to judge which possible moves would most likely be beneficial by percentage if the AI was able to "think" or calculate at least a couple moves ahead by at least 3-5 moves.
@Phoenixspin 7 років тому
It's over for us humans. We're done. Stick a fork in us.
@RahxephonXtra 8 років тому
@JustYourAverageRetro I hope this doesn't become a thing now. Everyone wants a bit of the pie. Those likes belong to the legend who has been doing it for weeks/months/years/decades. -Never forget 2016
@keatonlarson4657 8 років тому
So after the tournament they need to have 2 of those playing each other nonstop just to see how advanced it can get.
@sul141 8 років тому
Looks like the developers were inspired by Stephan Zweig's famous novella "The Royal Game".
@zachcrawford5 8 років тому
Google is selling Boston Dynamics. I'm worried that microsoft, GE or [shudders] Facebook will buy it.
@TheFishCostume 8 років тому
+Zach Crawford Microsoft might not be too bad, but Facebook would be a disaster.
@vladnovetschi 8 років тому
+Zach Crawford me 2 m8
@vladnovetschi 8 років тому
+TheFishCostume facebook will be a disaster, unless google buys it, microsoft wont be bad but neither will it be good, but apple, if apple buys boston dynamics... it will be better for us to suicide than to live to see what happens after aplepocaplypse
@MIQofDMC 8 років тому
+Zach Crawford There were rumors that Toyota and Amazon were interested in buying it.
@vladnovetschi 8 років тому
toyota should be the most interested, amazon should worry about other things
@nappybiscuit 8 років тому
Sounds like a lot of code to write.
@Roxor128 8 років тому
+nappybiscuit Not necessarily. It might be a neural network system, which is comparatively compact from a programmer's perspective.
@BatteryAcid1103 8 років тому
+Roxor128 That's exactly what it is. Although, it's an extremely sophisticated nn, with all sorts of stuff specifically tailored to help AlphaGo understand certain values in terms of how "good" or "bad" a certain move is, as Hanky Poo touched on in this video.
@CookiePepper 8 років тому
+nappybiscuit Like you do not use code for your brain, there is no code in neural network (except neuron emulation and learning).
@waynemv 8 років тому
+Acid, I expect the outer code that invokes the neural net was programmed with all the rules of the game, along with special code for doing basic tree searches, dividing computations among multiple processors, and basic stuff like that. I expect it would be able to give feedback to the neural nets when they've suggested an illegal move (but no longer needs to since the neural nets know the rules now.) But the neural nets themselves still had to learn everything about the game from scratch in order to be able to do well at evaluating board positions. Absolutely nothing in how they evaluates positions was ever hardcoded. Even the very very basic rule that players have to take turns they had to learn from examples, along with the rule for how stones are captured, that one can't recapture in ko, how the game is scored, etc. AlphaGo's ability to manage time is controlled by yet another self-learning neural net, which also had to learn from experience how to manage time optimally and to work alongside the other neural nets to determine when they can best benefit from having more time or when giving them more time would just go to waste.
@BatteryAcid1103 8 років тому
***** Most of what you said makes sense, but imo all of that code is still part of AlphaGo as a whole. It may not be a part of AlphaGo's brain, but it's a part of its environment, if you will. However, that bit about learning time, I'm relatively certain isn't entirely true, mainly from listening to what the engineers have said about the infamous move in game 4. Not 100% sure, but I'm fairly certain that the aspect of time and deciding when to stop searching for optimal branches it hardcoded, more or less.
@tobytalks2721 8 років тому
Wait... So its a bit like how bumblebee talks in transformers? By seeing how people respond from prior experiences, and the. Re-applying them into actual situations?
@greghuffman3061 28 днів тому
i got here by looking up ai vs ai matches for video games lolz. above this were like marvel vs capcom matches and starcraft 2
@FelipeL67 8 років тому
Okay Hank, now make scrips to beat Faker and not get banned like Dopa ~Kappa
@beshoytewfik895 8 років тому
Thank you for making me giggle
@kermanguy1877 8 років тому
But can you make it fun to play against, instead of just impossible?
@Teth47 8 років тому
+Kerman Guy Sure, why not?
@speedy01247 8 років тому
+Kerman Guy it already is, think about how big a challenge this computer is, and the fact is that it can still lose, so it is not impossible.
@Joeyyukonm 8 років тому
Everyone plays strategy games at different expertise. "Fun" would be extremely subjective.
@overwrite_oversweet 8 років тому
Just shorten the search tree? Heck, we could probably make a web app with it. Automatically throttle for low ranked players and give tips for learners.
@colinm.3419 2 місяці тому
:53 no! The goal is to get more points than the opponent, not to maximize the amount of points achieved. This discussion topic came up in the live commentary and DeepMind AlphaGo documentary- it is normal for humans to use the quantity of space captured as a proxy for how well they are doing, but AlphaGo only cares about preserving the lead, no matter how slim. This is why it was observed placing 'slack' moves. It already 'knew' it only needed to preserve the lead, making a decision not to increase it.
@folmez 8 років тому
Sedol was not trying to confuse with the wedge. It was simply the best move he had though only he could see it during a game. AlphaGO calculated the probability of the wedge being played 1/10000. That's why it didn't give it much thought. If it had, it would have responded differently because in fact the wedge doesn't guarantee Sedol a win.
@user-dk7xe1mz5i 8 років тому
Says spoiler alert than spoils in the next second. Like wot?
@SpaceNavy90 8 років тому
"AlphaGo might not be the best Go player in the world right now" he says after the computer won 4/5 games against the WORLD CHAMPION. Yes, it is the best Go player in the world right now. Jesus christ.
@rich1051414 8 років тому
+SpaceNavy90 It's debatable. He is definitely 'arguably' the best Go player, but at worst, he is one of the worlds best, and definitely competitive. 4th game in, the world champion learned a way to trip up the AI, so its not infallible. It is possible a human can beat him right now, with the right strategy. But it does not look good for humans at this rate :P Before it can officially be the worlds best, it needs to play through an ENTIRE tournament, and see if actually can win it. Only beating the best, and skipping everything that would normally lead to that point, is not a fair comparison, as beating a single person does not make you the best, even if that person is the best.
@jimm60701 8 років тому
+SpaceNavy90 if you watched the last video on the topic you'd understand why winning one tournament isn't actually all that big of a deal in go. it would have to win a few more before it could take that title.
@pyr0bee 8 років тому
Top ranked Go players are very close in terms of skill level beating him in 1 tournament doesn't make it the best. Personally I would like to see AlphaGo play against Lin Ke Jie because he beat Sedol this year. It's also interesting to see if AlphaGo can win multiple tournaments consistently
@xThirdOpsx 8 років тому
So interesting.
@timothy2214 8 років тому
"More possible games than atoms in the universe" Heh I learned that from Terminator... which is all I can think about after hearing the mention of cybernetic revolutions...
Fabiosa Best Lifehacks
Переглядів 9 млн
Суспільне Івано-Франківськ
Переглядів 598 тис.
Neat AI
Переглядів 515 тис.