AlphaGo vs Ke Jie, Game Three

Also, Ke Jie requested to play as white in this game as he cannot win the overall match anymore.

>>60602229
I'm rooting for AlphaGo.

Will we see an influx of band-wagoners and low level players filling this thread with their nonsense again?

>>60602941
Typical Ke Jie, he loves his white stones.

30 minutes

File: 10801668.jpg (190KB, 1920x1080px) Image search: [Google]

190KB, 1920x1080px

Pre-show is on.

I read DeepMind is looking into tackling Starcraft next. True?

When will Google create Go and Chess software in search like Solitare, Tic-Tac-Toe, Random Number Generator, Spinner, etc? I'd like to play Tic-Tac-Toe against Google DeepMind.

File: 14212633762.jpg (222KB, 1920x1080px) Image search: [Google]

222KB, 1920x1080px

>alphago gets no food
Fuck the chinks.

It begins!

>>60603765
Shitty microphones, audio mixing, camera angles, shitty interpreter...
The Chinese Go organization is either incompetent or trying to undermine this event.

File: 145313262119397360.jpg (398KB, 600x600px) Image search: [Google]

398KB, 600x600px

>>60603765
it's already fed by the tears of Chinese coal miner orphans

>>60603765
Google DeepMind's human may not ingest his tea.

>>60603809
She's sexy.

Looks like google rigged alphago to take longer with its moves than in the previous games (45s -> 60s). In reality, it probably needs just few seconds.

>>60603813
That's because the chinese have drugged it to try to make AlphaGo do a mistake.

BOOORINGGGGG

should just bring out like 5 top chinese go players and have them collaborate against alphago.

>>60603902
They did that yesterday. Group vs AlphaGo.

File: muXx1kp.jpg (359KB, 1100x3300px) Image search: [Google]

359KB, 1100x3300px

>>60602229
who /math/ here?

https://jeremykun.com/2013/02/08/why-there-is-no-hitchhikers-guide-to-mathematics-for-programmers/

>>60603804
it's held in a backwater town. google is officially unwelcome in china, that's why

>>60603935
i don't take life advice from bloggers who communicate with impact font images

>>60603935
decent article

kind of sick of the machine learning meme desu, literally the least interesting area in math and programming. people who deal with it are just feeding a black box data and aren't much better than robots themselves

>>60603765
>>60603804
AlphaGo should play in Japan as well.

>>60604120
They don't have any good players.

>>60604125
w-what about that one young guy, ichiro or something, he's promising isn't he?

>>60604143
No.

Ke lost already.

>>60604449
hello Nie Weiping

>>60602229
>3 rounds
>AI is 2-0
>Still play final round
Humans need to be replaced by robots once and for all.

>>60604485
Google paid him to do all 3 rounds.

Something about "muh data"

They should hold a one month event where all the pros have a chance to play against Alphago. Have the pros hammer away at Alphago 24/7 in shifts, playing together or alone, throwing everything they have at Aphago. You can join the event if you have X amount of dans.

Beating Alphago just once earns you 1M DOLLARS prize money.

>>60604514
That's pretty much what happened last year. It beat 60 pros in a row

>>60604514
I'm more interested in seeing how some of the other 10d bots compare to AlphaGo.

File: akot.jpg (580KB, 1200x628px) Image search: [Google]

580KB, 1200x628px

>>60603827
Is kot male or female?

AlohaGo will be so specialized that could be beaten by noobs using unconventional moves after a year or two

>>60604548
eta koshka

>>60604557
i don't think the hawaiians are that far along though

>>60604514
I thought it did lose online sometimes, back when it was called something else and wasn't known to be an AI.

>>60604568
Hawaii will be a superpower by 2040

>>60604573
AlphaGo has never lost, though there was a timeout or cancelled game or something when it was playing as Master on Fox.

File: qkot.jpg (25KB, 250x201px) Image search: [Google]

25KB, 250x201px

>>60604568
What?

>>60604607
этo кoшкa, ты дypaк

File: china.png (16KB, 800x533px) Image search: [Google]

16KB, 800x533px

China is so upset after Ke losing first match that they banned streaming of all future matches.

https://www.theguardian.com/technology/2017/may/24/china-censored-googles-alphago-match-against-worlds-best-go-player

> It appears the government was concerned that 19-year-old Ke, who lost the first of three scheduled games by a razor-thin half-point margin, might have suffered a more damaging defeat that would hurt the national pride of a state which holds Go close to its heart.

> After the game Ke said AlphaGo had become too strong for humans. “I feel like his game is more and more like the ‘Go god,’” he said. “Really, it is brilliant”.

> The ban underscores the esteem in which Go is held across east Asia, where it has been played in more or less unmodified form for over 2,000 years. First invented in China in 500BC, it was considered one of the four arts a scholarly Chinese gentleman should master, along with playing the guqin, calligraphy and painting.

>>60604626
>China is so upset after Ke losing first match that they banned streaming of all future matches.
Read your own article, it was banned from the beginning.

File: 1495568196736.jpg (198KB, 1000x667px) Image search: [Google]

198KB, 1000x667px

>>60604625
rude

>>60604654

I mean Ke already lost to AlphaGo in January, during 3 games.

>>60604589
It was beaten by the Korean once.

>>60604696
you caught me

>>60602229
Can someone give me a primer on Go?

What's the allure?

I'm a chess player and Go never grabbed my attention. Even though it's a complex game, it seems to lack the dynamism that pieces with different abilities gives to chess.

Should I look more into the game or keep ignoring it the same way I ignore Yu-Gi-Oh?

>>60604724
less rules + larger board = more complicated game
i wouldn't call it less dynamic than chess, it's a different game
>what's the allure
try playing a few games or watching/reading Hikaru no Go, that'll give you a good idea

>>60604724
>dynamism
dynamism in go emerges from formations and the way they interact. there are different "types" of groups - weak groups, strong groups, thin, thick, bad shape, good, two eyes, one eyes, no eyes and running, dragons...the shapes are extremely diverse and the game proceeds from the interactions of these shapes

the key difference is that you and your opponent are responsible for which shapes emerge in a given game. you want to be strong, but your opponent is trying to make you weak. and because there are many groups on the board, every move has dual purposes. a player that is better than you will "gain" moves on you by playing flexible moves with many purposes, and then you will find yourself in a position where you have more weak shapes than is tenable, and something dies

if you've never played the game at all it is essentially impossible to read the board, because you won't "see" these shapes. you'll just see pieces and the game will look like the interactions of pieces. but in fact close-piece reading is only a small part of Go, the game ke gie is playing right now is being determined almost entirely by many-purpose moves and a struggle for life and strength

File: Ke Jie can not lose here.jpg (121KB, 1921x1078px) Image search: [Google]

121KB, 1921x1078px

>>60602941
>>60602996

>>60604724
Chess is to Go what Yu-Gi-Oh is to chess.

Difficulty through complex rules vs difficulty through exponential state space.

File: ke jie despair.jpg (46KB, 580x435px) Image search: [Google]

46KB, 580x435px

This picture says it all. Ke Jie hasn't resigned yet, but he can see where this is going. It is just heart breaking.

And we are talking about the current world champion. A YOUNG world champion, who should have many years left to lead the game community forward. But here he is, in despair.

>>60605248
why did you not just take a screenshot

>>60605248
GO IS DEAD

>>60605283
>why did you not just take a screenshot
It isn't mine. I am watching a different stream.

Would be interesting too see Strong AI integrated into daily software.

Like imagine, you want to do graphic design. You open an application and tell it exactly what you want. Poof, the logo/design/interface/wallpaper is made. Then you describe the changes, and it's done.

You want a program for a specific task, you just tell the system in general terms & it's made, ready to use.

You want a retro fps game in the style of doom using a caveman theme, a few minutes later the game is made ready to play.

We'll probably see strong AI integrated into the economic system as well. The invisible hand of free market is bullshit compared command style economy controlled by Strong AI. Unfortunately you would lose all privacy, but your fridge will be stocked with every item you want. Self driving cars will make deliveries in minutes. Higson bison type of discoveries on a daily basis.

>>60605346
>double-spaced, short paragraphs
>delusional thinking
>doesn't care about privacy
i know where you're from

>>60605316
No.
Chess got more popular than ever, because of chess bots.

>gook manlets taking their board games too seriously

I'm glad robots are taking over desu

>>60605417
And this also eventually will lead to better analysis of the game leading to better players.

>>60605380
>privacy
>gay cyberpunk future
pick one

>>60605346
how do you think AI would be able to do graphic design or program?
The AI has to be incredibly general to actually program but where would it be learning from? How would it even debug the application? Not even going to ask about the retro fps game.
Did you just think that AI can do whatever with no limitations? Did you even think before you typed any of what you typed?

>>60605439
False.
Current chess players are worse than they were 20 years ago, but the game is more popular than it was at that time.

how do I git gud in Go? I get the rules, but CPUs always beat me in harder difficulties, I just ladder myself up

>>60605493
No idea. I can't even beat the simplest bots out there.

>>60605346
You mean Higgs boson

>>60605346
pretty sure animation is already working on AI for tweeting, because that is some grueling slave workshop work and everyone wants it done faster

>>60604724
The move A Beautiful Mind had the main character playing Go

>>60605248
The late game is where AlphaGo really shines so if he doesn't have a confident lead by this point he may as well resign

>>60605546
>The late game is where AlphaGo really shines so if he doesn't have a confident lead by this point he may as well resign
At this point it isn't about win or lose. Ke Jie just doesn't want the game to end. He can't win anymore but he wanted to see it through.

This would be considered rude if he was playing a human, but Alphago thankfully couldn't be offended.

>>60604626
>painting
What do Chinese paintings look like?

#rekt

Christ these ugly chinks are so fucking boring.
Why can't they just get the white guy back?

>>60605671
Those girls are cute!

Well, it's over. As expected AlphaGo just btfo everyone in the event.

I will be welcoming our new robot overlords in the very near future.

>>60605792
>Well, it's over. As expected AlphaGo just btfo everyone in the event.
>I will be welcoming our new robot overlords in the very near future.
There is a silverlining for all the players in the world. Alphago can beat any human, but it is incapable of teaching humans why it does anything. So all the game instructors can keep their jobs because players still need good teachers to understand how to get better.

>>60605459
If the game is more popular now then are you sure you're not just dragging down the average?

File: 1495704401427.jpg (188KB, 730x864px) Image search: [Google]

188KB, 730x864px

>>60605248
>IT IS HAPPENING AGAIN

>FUN = WINNING

This guy is so butthurt, rofl

who is watching the post game interview?

>>60606061
Lee Sedol beat Alphago, which is something I guess.

File: Screenshot_20170527_103831.png (252KB, 533x441px) Image search: [Google]

252KB, 533x441px

>ruining a perfectly good goban with ugly scribbles

Has this happened yet? I can't find a link for it.
>Pair Go: human plus AlphaGo versus human plus AlphaGo

Hassabis confirmed that there will be no future matches between AlphaGo and professional players. This was the last.

>>60606061
Computers will teach humans.

>>60602229
>>>/v/

>>60606265
>be the best in your field
>get schooled by an AI
I dunno man, he might have enjoyed it if he was playing a human who was better than him. Having a computer take you apart so easily probably feels bad.

Thank fuck this shitty game no longer has to be the benchmark for AI.

>>60606488
What did Go do to you?

https://deepmind.com/research/alphago/alphago-vs-alphago-self-play-games/

10 games in which Alphago plays against itself.
A lot of these moves are pretty crazy.

>>60606469
life is cruel

>>60606061
>mfw robots are pretending to be dumb so they don't reveal their secrets to humans

>>60606721
8 of those games were won by white. I wonder if this is just a coincidence or if AlphaGo prefers white with this komi.

>>60604724
>Should I look more into the game or keep ignoring it the same way I ignore Yu-Gi-Oh?
please dont talk about yugioh again, its my trigger.

rip chink guy ;_;

>>60605346
>Higson bison
physicians get out reeeeee

>>60607662
Yu-Gi-Oh

So Deep Mind has released 10 new self-played games of Alphago, and will release more over the next few weeks until they realise 50 new games total.

The interesting thing with these games, is that players can barely understand what is going on. At least with a human opponent you can use Alphago's reactions to humans, to judge what it is doing. But with self play games Alphago just go ham and counter-counter-counters itself. In one game White Alphago sacrificed 30 points to win later. One of the human players reviewing didn't even realise that White won that game because he had assumed that the move was fatal.

>>60606469
Eh, it learnt from human games, so technically speaking you're still playing against humans. It's just that they're many humans at once and they're on electron based turbocrack

>>60606344
Computer-assisted go will probably become a thing, just like with chess.

>>60609382
>Eh, it learnt from human games, so technically speaking you're still playing against humans. It's just that they're many humans at once and they're on electron based turbocrack
It became less and less true after each upgrade. The latest version basically learned exclusively from old data of its own games.

File: 1494607671368.png (507KB, 588x617px) Image search: [Google]

507KB, 588x617px

>>60606320
This is the first thing I though of when I saw that, only it's the asians handing things over

>>60609367
Yeah AI is pretty fucking intense in that way, how far ahead of your thinking it is. It's going to fucked once it learns to get out of the boxes we put it in. And I think we're pretty close.

File: alphago pole.jpg (55KB, 580x579px) Image search: [Google]

55KB, 580x579px

>>60609367
Here is an example of an Alphago self play opening.

Look at that black position. All on one fricken line. That is literally alien to us.

>>60609610
Alphago really is a great example for what is often said about AIs.

AIs are usually not perfect. That is why Alphago still unpredictably win or lose against itself. If it was perfect the winner would be fixed every time.

But Alphago also once again proves the saying "The AI doesn't need to be perfect, AI just needs to be better than humans".

Alphago had not solved Go, not by a long shot. But since it already is better than any living human there is no need to argue that it is not perfection personified. It is "good enough" for our purposes.

>>60609610
did black win that game?

File: 1495821925819.jpg (86KB, 1024x1024px) Image search: [Google]

86KB, 1024x1024px

>>60609610
Moar pls. This shit fascinates me to no end.

Also, anyone have any YouTube video links reviewing these self play games?

>>60609910
>did black win that game?
White won.

But that matters less than to think that such an opening i even playable.

Here is the link to the games. The one we are talking about is the 2nd self play game.
https://deepmind.com/research/alphago/alphago-vs-alphago-self-play-games/
I find that I can't use Firefox to run the games, I had to use Chrome.

>>60609950
>Moar pls. This shit fascinates me to no end.
Deep Mind will be releasing 10 games every day until they released 50 total. More than enough to keep the Go community busy for a year.

>>60610011
I meant screenshots and you talking about its moves :3

>>60610062
>I meant screenshots and you talking about its moves :3
I am sure there will be actual professional teachers uploading analysis on Youtube for the next weeks and months. Leave it to the pros.

>>60609367
>One of the human players reviewing didn't even realise that White won that game because he had assumed that the move was fatal.

Link to this game? Or video of that guy reviewing it?

>>60606061
Did you see the latest blog post?

>We’re also working on a teaching tool - one of the top requests we’ve received throughout this week. The tool will show AlphaGo’s analysis of Go positions, providing an insight into how the program thinks, and hopefully giving all players and fans the opportunity to see the game through the lens of AlphaGo. We’re particularly honoured that our first collaborator in this effort will be the great Ke Jie, who has agreed to work with us on a study of his match with AlphaGo. We’re excited to hear his insights into these amazing games, and to have the chance to share some of AlphaGo’s own analysis too.

>>60610155
>Link to this game? Or video of that guy reviewing it?
Sadly it is comments from a go forum. It is once again the 2nd self play game. Remember that it takes time to go through ten games so the 2nd one just got more attention paid to it than the 8 others after it.

Just wait a few days for real reviews.

>>60610233
>Did you see the latest blog post?
Yeah, but the teaching tool still need humans.

It's like how you really can't be a pro chess player just from training with a computer, even though computers are stronger than humans already.

File: eaa60cf431adcbefc44a357ba6af2edda3cc9f51.jpg (30KB, 580x388px) Image search: [Google]

30KB, 580x388px

human.

>>60610280
Ah yes, right after AlphaGo generously gave the humans a bunch of pity points

>>60610358
After looking at the self play games, it is pretty clear that Alphago only play as well as it needed to.

I still remember last year when a commentator said that Alphago's moves against Lee Sedol was rather simple to comprehend even if it was not easy to predict.

But new version aside, the new self play games imply that Alphago would go to extreme lengths to win IF it is being pushed by someone of equal ability. That the simpler Alphago plays, the more likely that Alphago is metaphorically cruising to victory while half asleep.

The team match was never going to be better than having just one player. And it appears Alphago ended up beating them with ease because of this.

>>60610242
>a go forum

It's okay. You can say Reddit ;)

>Releasing only 50 games worth of self play data

Google, you have like thousands of these games. Even like a set of a thousand doesn't take that much more effort. Just let the go experts analyze that set.

On the other hand, I'm sad that Google isn't setting up a cloud instance or something for AlphaGo to permanently live so people can play it from around the world. They could charge for it and it wouldn't be horrible.

Kinda a shitty way to go because Deepmind could've done a lot better with leaving Go and announcing their closure but meh, I guess Starcraft or something will be next.

>>60611674
Top players can't even comprehend a single of these games. 50 of them will puzzle players for decades

>>60604102
>implying black boxes that only computers understand aren't the shit

>>60611674
>On the other hand, I'm sad that Google isn't setting up a cloud instance or something for AlphaGo to permanently live so people can play it from around the world. They could charge for it and it wouldn't be horrible.
They are planning to release some sort of a teaching tool, and Ke Jie agreed to help them with the tool's development. So some sort of Alphago analysis software will probably be made. I don't think a Cloud Alphago is out of the question, it is just a matter of what is the most practical.

It also depends on how much their final Paper on AI would reveal. It might reveal very little, or tell us enough to build Alphago from scratch.

>>60612325

Judging from their Nature paper, unless they are going to go into more detail, it's not going to be enough to build it from scratch. They might just give out at least the architectural details without the drawing plans of the improvements and then leave it at that.

But apparently, Wired and Verge said there were "issues" regarding a wider release of AlphaGo which was why they couldn't let it live in the cloud.

File: elo.jpg (43KB, 800x462px) Image search: [Google]

43KB, 800x462px

Daily reminder that no human ever stood a chance.

This is like playing arm-wrestle against a forklift

>>60613017
The scary part is that Alphago basically imitated how humans learn the game, but then was able to train itself much faster than any living human. It is the opposite of how AIs usually brute force things. Instead of brute forcing a single solution, Alphago brute force EXPERIENCE instead.

Alphago simply played more go games than quite possibility the rest of humanity, living AND dead, combined. And it doe this by playing against other versions of itself, resolving a full game in seconds and start a new one, over and over.

Alphago didn't beat humans by out-thinking us. Alphago just is the the most experienced Go playing entity ever. it knows more about the game than humans outright.

Humans are officially obselete.

>>60613017
>>60614612
People in this thread forgot to mention that DeepMind also made an Anti Alpha Go, which is basically Alpha Go's evil brother who plays ridiculous moves on purpose, to help train Alpha Go against Go players who just want to shitpost while playing.

>>60615381
That's actually a really important fact.

Because despite what people in this thread will have you believe, AlphaGo does NOT know how to play go. Machine learning AIs are extremely weak against degenerate strategies just because it has never seen them before. You have to explicitly tell it how to play against an idiot even if it can beat top pros

>>60616663
>Because despite what people in this thread will have you believe, AlphaGo does NOT know how to play go.
<SATORI>
As soon as we stop insisting that our machines be anthropromorphic, the better the machines will be because we will see their true nature.

File: IMG_20170528_063402.jpg (26KB, 640x513px) Image search: [Google]

26KB, 640x513px

Is this what it feels like to lose your job to an AI?

>>60616663
>Machine learning AIs are extremely weak against degenerate strategies just because it has never seen them before.
That's the same for pros though.

>>60619338
He didn't lose his job, and he probably only got stronger from these

>>60619580
No it's not because they actually know how to play go.

Just like a black belt doesn't need to see how to beat an idiot flailing his arms to walk up and kick his head off.

>>60604102
I like the results it can produce tho

>>60621236
It's common for black belt "martial artists" to get served by street fighters because their fighting style is unconventional to them.

Your analogy is why you are wrong.

>>60603765
Alpha Go is powered off the tears of human Go players

>>60621934
>It's common for black belt "martial artists" to get served by street fighters because their fighting style is unconventional to them

This is what happens when people watch too much anime.

Anyway, a competent yet unconventional style is much different than just doing stupid random things.

Example, there is a deep learning bot that can beat pros at smash bros. But one of the pros figured out if he crouched next to the edge then the bot would go crazy then eventually suicide. This was because it was simply never trained against all possible inputs

https://www.youtube.com/watch?v=OCevCII1zo0

They published the second batch of ten AG self matches.

https://deepmind.com/research/alphago/alphago-vs-alphago-self-play-games/

I want to code an aggressive/kamikaze chess engine that favors sacrifices instead of optimal play to rekt kids in online chess and laugh alone like the autist I am. I already did a bit of research about how chess engines works, I need to code the board representation, the search function, the evaluate function, etc... But do you have any idea how to make it focus on sacrifices that it can recover and wreck the opponent some moves later? Maybe give a higher score for moves that evaluate low for a few turns, then suddenly evaluate high?

>>60622345
Yes, competitive martial artist, believe everybody would play nice,no tried kill it and only fight until surrender, i.e. taekwondo use only kicks and tried to score more points,don't expect close attacks.

Other hand muay thai fighter would fuck it so hard.

>>60624314
You will need to go deep in your alpha-beta

>>60624087
Scrap that, looks like they put all 50 out by mistake.

https://s3-us-west-1.amazonaws.com/apetresc-site/alphago_selfplay.zip

Wins were 38-12 for white.

Another weird sequence from an AG v AG game.

File: DA68FcuXYAEuJvz.jpg-large.jpg (36KB, 353x354px) Image search: [Google]

36KB, 353x354px

>>60626159

>>60603883
what are you even doing on /g/, you tech illiterate normie ?

>>60604485
>what is sportsmanship

Sure, if it were a private event that would be one thing. But because it is publicly viewed by a rather large audience, gestures like playing all three matches mean a lot to those who would be entertained by this. For the competitors, not much is lost since they agreed to playing up to 3 full matches anyway.

>>60624519
Surely you understand the difference between a muay thai fighter and a guy that just crouches by a cliff