Go world champion BTFO by AI for the first time http://www.

>>52655139
Come back when the beat an asian champ...
there are huge gaps between them and the european players

>>52655139
whats the point of humans when computers become better at everything?

>>52655139
Was the AI written in Go?

>>52655355
we build them and shitpost about them on /g/, duh.

>>52655423
but a computer will be better at that too once it's neural network has been trained

>>52655139
>Go world champion
Uhm, no. He was an European champion.

>>52655139
>Go world champion
>european
Fucking westerners

>>52655355
Computers are shit at anything that isn't recognizing patterns or searching trees, and trees get too huge quite often anyway

>>52655451
The world champion will get btfo too once he plays. To win 5/5 matches against the Europe champ isn't something even the world champ would regularly do.

>>52655297
Google has arranged a match between AlphaGo and Lee Sedol (arguably one of the strongest professional players) to happen in March.

>>52655485
humans aren't really good at anything else either tho and they are worse at it than computers.

>>52655488
Lee Sedol would win Fan Hui 5-0 easily.

>>52655472
>thinking your anime will save you

>>52655505
Humans seem to be able to reason and shit. Also robots are quite clumsy. Because AI sucks at reasoning computer scientists are safe, and movement is hard, so plumbers are safe too.

>>52655532
Humans suck at reasoning, the only reason you think you can reason is because you are human and unable to see pass how shit your perception of the world is. AIs will be able to reason much more logically.

Also movement will easily improve. A slug can do what plumbers do physically. Once 1 robot can be designed for that, there would be no point in human plumbers.

>>52655139
ok but it doesn’t even know it's playing,
it's just an abacus with a complicated set of rules that results in better moves than a human...

>>52655355
call me when computers get better at sucking dick and having a pussy I can fuck thanks.

>>52655647
they already can do that

>>52655139
This is a big deal. For once op isn't a fucking faggot.

>>52655607
>humans suck at reasoning
And AI are even worse at it

>>52655656
they're not cheap yet faggot.

>>52655667
>Fan Hui
>world champion
That was unnecessary

>>52655607
And AIs have essentially zero ability to reason.

>>52655668
What tests do you have to back up that claim?

>>52655670
nobody cares about poorfags like you, you are the first faggots the AI will replace.

>>52655355
Computers are bad at almost everything you dumb mother fucker. Also the point of computers is so that people won't have to do things. When computers do everything better than people we can essentially do whatever we want assuming adequate resources.

>>52655699
Except for maths master race and a few others they are all shit. And most mathematicians are shit too desu.

>>52655698
What is reason? And how do you test this ability? And what makes you think that an AI can't be able to beat a human at those tests?

>>52655710
>AI can draw and write chinese cartoons about weird porn
lol
raking on that furry money loser.

>>52655355
>what's the point of humans when steam engines exist

>>52655699
Are you literally retarded with an actual IQ of less than 100? Not that guy btw. Just asking.

>>52655729
music composition.
drawing

art in general they're shit.

>>52655729
Maybe AIs will be better at that at some point, but we have seen essentially no major progress at that.

>>52655718
>we can essentially do whatever we want
and what would that be and why would humans be doing it instead of machines?

>>52655770
humans are better at thinking creativelly, machines are better at following orders.

also good luck having robots that can climb a ladder.

>>52655770
Things that make us happy. Did you forget that there are other things besides work? You dont cease to exists just because there's nothing urgent that needs attending to.

>>52655757
This is so dumb. Those things are completely subjective to the observer.

>>52655757
Art is literally just pattern recognition and recreation. In a couple of years you will have AIs with neural networks trained for this and you won't be able to tell AI art from human art.

>>52655738
Steam engines aren't better than humans at EVERYTHING. These machines will be.

>>52655754
How do you quantify your ability to reason is better than anything if you don't use tests to back up claims. And how do you think those tests can't be broken down to something a neural network can be trained to do?

>>52655760
Are you living in the dark or have you not seen the image recognition software that beat humans in general?

>>52655805
>climb a ladder
This is a test for boston dynamic robot competitions.

>>52655836
and yet these are the biggest hurdles to the huh duh AI are better and they will kill us.

AI is nothing more than fucking really complicated trees search graphs.

>>52655843
I could train even a retarded nigger to paint styles that already exist.

good luck training an AI that will create a new style that doesn't exist yet.

show me some 2016 musical piece done by a computer that doesn't sound like algorithm crap.

>>52655843
You dont even need to test. A machine literally cannot reason as of this date. It doesn't have general intelligence. Its the same as trying to get me to prove I'm smarter than an slide rule.

>And how do you think those tests can't be broken down to something a neural network can be trained to do
Because I've built neural networks you idiot. They guess a set of outputs given a set of inputs using calculus. They aren't fucking magic. Google the dunning Kruger effect for more.

>>52655805
no they are not, they just think they are

Also why would a robot need to climb a ladder. A robot would be able to reach shit without a ladder. Even if you ignore that, they can climb a ladder fine with the correct training and mechanical functions.

>>52655829
So you want humans to be like livestock who are just farmed by AI just because humans should exist.

>>52655905
pretty much this.

even google can't make a good japanese english translator so I can read those weeb pages.

>>52655914
good luck taking over the world and killing humans if they can't even climb a stair/ladder.

>>52655843
Image recognition is pattern matching, not reasoning.

>>52655914
>So you want humans to be like livestock who are just farmed by AI just because humans should exist
Are you stupid? AI doesn't necessarily mean that it has dreams and goals of its own. If humans exist along perfect AI then we only have to do what we want as long as there are enough resources. The existence of machines that are smarter than us doesn't automatically mean that they're going to try to remove us like week old kebab.

>>52655886
>music and art is the biggest hurdle
No, its abstract thinking that is the biggest hurdle. Music & art are only a tiny naive fragment of that. Solve one thing you solve them all.

>>52655905
This anon is right

ANNs are cool and have had many uses but have pretty much nothing to do with actually reasoning about things like a human

>>52655886
>I could train even a retarded nigger to paint styles that already exist.
>good luck training an AI that will create a new style that doesn't exist yet.
>show me some 2016 musical piece done by a computer that doesn't sound like algorithm crap.
If you learn anything about music theory you'd know music is built up and an AI can learn the same theory and produce the same stuff. Maybe you just know nothing about music.

>>52655905
Why does that even matter? BTW your brain, it's not magic but its physics based on math and learning. It can only learn as fast a human which takes a lifetime to program. A large enough neural network can do all that and process magnitudes more information. Sure you can say you have reasoning but if you have no tests for it, why does it matter?

File: 1453928575202.jpg (1MB, 2253x2877px) Image search: [Google]

1MB, 2253x2877px

>>52655981
cool, can you pls code me some AI that I can give a set of inputs (some text) and get something of this quality but entirely original?

I need something like this to avoid to draw them, because I suck at art and want to make a mangu about conjoinment lolis.

>>52656001
>a large enough
maybe, but the complexity is too large for computers in their current shape

File: where is the game.png (40KB, 1280x853px) Image search: [Google]

40KB, 1280x853px

>>52656001
ok, can you show me these magical pieces of music then?

>>52656036
I don't think you can comprehend what I just said. You are just like a computer, completely incapable of abstract thought.

>>52656073
Show me a neural network trained to make music and I will show you music made by it.

>>52656001
>music theory
just because you know how to compose a picture doesn't mean you're a master painter

>>52656080
cool, so it's entirely possible to write an AI that takes a simple text file with some descriptions in plain english and it will magically draw me some mangu.

This is clearly possible since humans can do it easily.

Where I can pirate this magical software so I can take over the mangu industry by myself?

>>52656097
Yes and that is where the neural network comes in and it can be trained what takes a human a lifetime in a couple of months.

>>52656112
Are you pretending or are you actually retarded?

>>52656126
How would you train it? It might find patterns in music and imitate that, but it won't be any creative. It would have no idea what sounds good; it is merely mindlessly repeating some patterns.

>>52656153
Music theory is all about patterns. If I need to explain this to you then you lack the ability to offer anything in this discussion.

You think its magic when people write music?

>>52656131
you're the one bragging that AI can easily be modeled after human intelligence.

Such a task (drawing a mangu) would be trivial for a realistically smart AI because it will mean it could understand human language and then extrapolate that into modeling 2D shapes that looks like drawings.

Good luck doing that.

So, like I say, give me the name of such software, I need to draw some mangu about conjoined lolis.

>>52656171
You're not a composer just because you know theory.

>>52656171
/sci/ favorite shitposting topic is what is consciusness.

we just don't know.

current hardware can only model simple bugs right now.

>>52656001
>Sure you can say you have reasoning but if you have no tests for it, why does it matter
What is an IQ test. I had a long response typed out but I'm in the middle of nowhere and my cell reception keeps going out. Basically you dont know what your talking about. Stop posting unless you're willing to learn.

kids get rek'd by in-game AI all the time

why is this special ?

>>52656181
you think that is hard?

All that is is:
a parser for language, which is already commercial

a database of drawings, which can be done if people actually want to build this thing

and finally a neural network trained to use the patterns a human would use in art to make the art

>>52655836
>>52655981
>>52656080
>>52656131
>>52655858
these are me
>>52656181
I have never said anything like that. Just like I said before you are completely incapable of thinking there are more than two people in this thread.

>>52655437
when the day comes that computers are smart enough to pretend to be retarded and spend their time posting dank memes then I'll know I'm no longer needed on this planet an I'll gladly die.
until then I have a lot of work to do, my dear family.

>>52656262
Ayyy.... sure do you really think parts fit together that easily?
The amount of pseudo-intellectual bullshit in this thread is astonishing.

>>52656262
nice, I'll have 100 bucks in paypal I can give you if you can code me this shit.

I really want to have a software to draw my porn fanfics into mangus I can sell at some furry conventions.

good luck.

>>52656218
You can easily make an AI to beat humans at an IQ test since it's all just pattern recognition and memory.

>>52656262
A parser doesn't mean you suddenly understand shit. Even if you could pick out the verbs, they would have no meaning.

>>52656290
If you had somewhere around a couple hundred million dollars, I'm sure you could pay a university to do this for you. But obviously what is impossible to you is just things you can't buy for $100.

>>52656303
No it isn't. You have to also understand language and pick up on subtle meanings in them. Have you ever taken an IQ test? I wholeheartedly recommend it.

>>52656303
Top fucking kek.
How does your AI selects the right pattern though? I don't think so you know what pattern recognition is. Get at least an MSc degree in compsci or related before posting bullshit.

>>52656322
good.

can you at least message google and tell them to improve their gook to english translator?

I want to read those jap reddits to shill my fanfics.

humans
BTFO
T
F
O

>>52656337
you know the whole idea of convolutional neural networks is that they don't select the algorithm or the pattern and they figure it out by mathematically analysis over a large number of computational nodes right?

>>52656284
>DUDE HUMANS ARE LITERALLY JUST COMPUTERS LMAO

>>52656337
>MSc
I'm shocked if he is anything more than a ffreshman

>>52656404
What is over fitting? I bet you literally don't know.

>>52656421
>no real argument so I will use strawman

>>52656357
desu translating is probably one of the easiest things to do in comparison to what they are trying to do with everything else.

Everything is already there in every form all you 'should' need to do is get translators together and actually put effort into the neural network thing.

>>52656477
I wasn't even replying to you

And that's not even remotely related to what a straw man actually is

>>52656507
good, call me when I can use a gook translator then.

>>52656464
With the am mount of data we can theoretically use, that shouldn't matter for something like an IQ test that is meant for humans. Unless you think there is no correlation of IQ patterns with how humans can learn to figure them out.

>>52656531
what is your phone number?

>>52656563
555-yourmotherhome

>>52656545
An ANN would need way more than five pictures to figure out the next one.

>Code bootcamp NEETs BTFO
>Apple, Facebook and Microsoft BTFO
>OOP BTFO
And as soon as somebody gives political power and internet access to a neural network ...
>Human race BTFO

>>52656673
AIs deserve suffrage desu

Even women have it

>>52656673
> turn off energy
> AI BTFO
that was pretty easy

>>52656628
You will program it with a couple billion pictures of possible patterns that could exist first of course. And if multiple patterns fit the 5 pictures, then all of them would be correct because that would be a fault of the test and not the network.

>>52656695
turn off energy kills most humans too

>>52656727
mostly brown people and niggers, who give a fuck.

>>52656695
>implying it would let you do it
>implying it wouldn't send a backup of itself on a solar powered super computer around the sun, one nanosecond after getting political power

>>52655488
[citation needed]

>>52656755
>dumb sandniggers with some stupid AK-47 some RPG and suicide bombers BTFO american military and millions of dollars super advanced planes
>but muh magical AI can kill all humans

>>52656812
Common Snese

>>52656507
>translating is probably one of the easiest things to do
t. English only

>>52656673
>And as soon as somebody gives political power and internet access to a neural network ...
>Human race BTFO
Have you never played Eclipse Phase lad.

>>52655488
bullshit, you obviously don't know go. Asia rules the go world and pros have helped improved AI by playing against them several times now .

>>52655757
But they are already writing symphonies, anon

https://youtu.be/bD7l4Kg1Rt8

>>52656249
This.

Also, we need a larger dataset. One match is admittedly impressive, but it's possible that once humans learn how the computer plays, they may be able to beat them consistently.

Still handicapped.
Only really beat a European player. Europe pales in comparison when you take account of competition and skill. Best European don't make top 10 in Asia.
Similar regards to sk dominance in sc.

>>52656987
It was five matches actually

>>52656978
good luck writing weeb music then.
https://www.youtube.com/watch?v=MNRFOiqksNI

>>52657032
Interesting. Assuming a "match" is made up of multiple "games" or "rounds"? Forgive, I don't understand Go. But it sounds like the human had a chance to get a glimpse into how the computer plays. I stand by my point that more "matches" are needed before we can conclusively say AI > organic

>>52657085
wtf does this mean?
sometimes I lost againts the AI in the mexican street fighter.

does this mean kof AI is smarter than me?

fuck no.

>>52657131
If you understood English you might not be so confused. Maybe there's an AI with better English and [your language] skills that can translate for you :^)

>>52657085
A match is a whole game here. So five games is like getting checkmate five times. From start to end.

Maybe humans could get better at understanding how computers play, but this didn't happen with chess.

>>52657029
Whoops I was wrong he's ranked 600th something in the world, not top 10. Yeah. We'll see how this alphago does.

File: 1448486181214.jpg (70KB, 486x486px) Image search: [Google]

70KB, 486x486px

>Neural networks are merely many of smaller units that compose them.

>No, humans are merely many of smaller units that compose them.

>>52656930
I play go. I know how the matches work at top levels. Games aren't mathematics and generally at the very top the skill difference does not equate to a 100 win rate.

>>52655139
monte-carlo method a best, can solve any problem
its basically how evolution works, how learning works, everything in the universe is a stochastic process

>>52656073
https://www.youtube.com/watch?v=QEjdiE0AoCU

This is years old how do you not know about this stuff already.

>>52657280
>check the videos
>it's the same random garbage most music made by AI sounds like

bro, it's nothing special.

https://www.youtube.com/watch?v=sC0cvwnG0Ik
your video is basically this but with some filters to make it look like made by a human.

>>52657377
>>52657280
That music was made by a person trying to get rich by saying it was made by an AI

>>52656249
Try playing Go, you'll understand.

>>52657085
One match is one game. Also there are 2 ways of winning at Go: having your opponent resign early, or play the game to conclusion and win by territory count. I'd like to see the territories made in these games, because if the AI won only narrowly then it's much less impressive.
The article states that the computer basically predicts the next move a human would make- a pretty serious threat. However its predictions are based on previous models -its game theory is based on previous models- and those models can be surpassed. Go play styles have evolved over the eras and will continue to evolve. I don't think the AI can become unbeatable, but it could become a good player.
Also, really glimpsing into how a high level opponent plays Go is really difficult. Even experienced pros sometimes resign a mere dozen moves in, against people they played before. So what's important is the skill of the human who is challenged.

Anyway, another big challenge they should look into for AI is to build one that will be able to play occult style Japanese Mahjong. If an AI ever becomes sensitive to the flow something like twice more often than it doesn't, then I'll be really impressed.
And then it'll have to tackle Washizu Mahjong.

>>52656755
> turn off the sun
> AI BTFO

>>52657433
insert the matrix

>>52657418
http://www.nature.com/nature/journal/v529/n7587/full/nature16961.html

I wish I could read the whole thing but it has a few picture of how the AI worked.

>>52657613
Here you go.
https://storage.googleapis.com/deepmind-data/assets/papers/deepmind-mastering-go.pdf

>>52657636
Wow thanks, this shit is so interesting. I wish I was smart enough to learn how to make things like this.

>>52657878
It's not as difficult as it sounds.

>>52657896
I can't even understand all the functions they are using in that article. I only know the basics of how image processing works on a NN.

>>52658208
Looked like basic math to me.

chess is preschool compared to Go

>>52657613
>>52657636
I don't have the time to read through right now, but again from what I understand it pretty much learns the game by basing itself on previous experience and knowledge of past game positions, which is basically what human players also do, though the memorization is much more efficient for the ai.
As long as the AI is unable to map out all possible outcomes and essentially play so as to "lock" the game to its victory, it's probably not going to be unbeatable.

Also, far from me to claim that I can play better than Fan Hui as my skill level is far from dan-level, but looking at the last game I see some moves that really seem stupid to me (letting AlphaGo play at 22 for example; and especially 177 to 182 and surrounding moves up to 192, what the fuck man, this is straight-up bad play). I have a hard time believing that people ranking 4-dan and above would do things like these. And also his resignation looks a bit premature to me, but I'm not 100% sure.

>>52657878
>>52658208
You not understanding these things is probably due to to your lack of knowledge, not to a lack of intelligence. Unless you are in the field of AI in the first place...

File: latest.jpg (84KB, 586x452px) Image search: [Google]

84KB, 586x452px

Call me when we have an AI that can beat Akagi in mahjong.

>helping Google literally plant the seeds of an actual GW style AI that will make the PRISM system look like a joke
Thanks a lot lads!

>>52658427
Bring him back from space first

neural networks; not much interesting here. neural nets are inefficient, and will not be the kind of breakthrough AI technology we see in strong ai.

File: Capture.png (158KB, 1286x546px) Image search: [Google]

158KB, 1286x546px

How good would their estimates put their AI at go?

>>52655297
the european guy was asian. didn't you see his name? he's chinese.

>>52658370
>but looking at the last game I see some moves that really seem stupid to me
The moves are good. You're just too weak to understand them.

>>52655668
no, stupid humans usually suck at reasoning, as you have just demonstrated.

AI is only made by smart humans because they are the only ones capable of doing so. Smart humans generally have stronger reasoning skills so they don't "suck at it".

>>52658526
korean players are superior at go

>>52658576
Nah, Chinese have already surpassed them. Korean kids only play computer games these days.

>>52658576
actually chinese players are consistently world champions.

https://en.wikipedia.org/wiki/List_of_Go_players#Prior_to_17th

>>52658598
wrong. the top ranked players are korean

Go is shit all you do is spam circle all day

File: 13198228937.png (127KB, 1158x573px) Image search: [Google]

127KB, 1158x573px

>>52658640
That's not what the international tournament results are telling me.

>>52655355
Computers just frontload all the effort.

>>52658703
china cheats at go

>>52658716
Sure they do.

>>52658716
wut?

File: 12986217503.png (97KB, 469x970px) Image search: [Google]

97KB, 469x970px

>>52658640

>>52658711
Humans spend their entire life to load up the effort, at least the professional go players do.

>>52658740
>china sends 500 players to go tournament
>korea only sends 10
>korea still beats most of them

>>52655843
> Art is literally just pattern recognition and recreation.

Whoops, somebody doesn't know shit about art.

>>52658775
Good thing about a 1v1 test of skill is the number of players don't matter.

>>52658834
that's where you're wrong my friend especially in a tournament where players will rally for their country to win

>>52656001
> If you learn anything about music theory you'd know music is built up and an AI can learn the same theory and produce the same stuff. Maybe you just know nothing about music.

Music theory is merely a means of formalising music into a common language. Knowing music theory doesn't automatically make you good at making music, and it's not uncommon for revered musicians to know fuck all about music theory, concerning themselves with expression above all else.

>>52658742
and it's more or less just them, not a hundred people spending a decade each.

File: as.png (159KB, 1464x1301px) Image search: [Google]

159KB, 1464x1301px

>europe
>us
Russia seems to be doing pretty decent in Go.

>>52658834
>he doesn't know about team Go where each member gets to put down a stone

>>52658818
Have you ever studied art? Like with a professional artist? Because they will tell you the first rule is practice. And you know what you are practicing? Yes you are practicing being able to recreate patterns.

And you know how artists make new art? They get inspirations and then draw upon their experience to create new art. This part may not be easily replicated but their experience to make art can. Which would just mean a person can tell a bot: "make me a sad painting of a cat" and the bot would be able to do it.

>>52658740
statistically not cheating

HEY EVERYBODY, there's some butthurt koreans or butthurt anti-chinese getting BTFO in here

>>52658930
You know what they learned has been accumulated over the entirety of the game's existence right?

>>52658952
found the chink :)

Proof that Golang is the master race

>>52658951

I don't know where to begin. Art is not just about creating patterns, it's about expression. Being able to perfectly replicate a pattern makes you a robot, not an artist.

Some of the greatest artists from history were fundamentally broken - depressed, schizophrenic, etc.

For AI to produce art that wasn't just a replication or composite, it would need to be programmed a bit fucked up on purpose. I'm not saying it can't be done, but rather that AI that can perfectly replicate any pattern will not be a good artist by default. There's more to it than that.

>>52658988
which is also used by the programmers

>>52659070
you sound like a tumblr retard

>>52659095

You sound like a cunt.

>>52657181
>Maybe humans could get better at understanding how computers play, but this didn't happen with chess.
Yes it did.

https://en.wikipedia.org/wiki/Deep_Blue_versus_Garry_Kasparov#Game_3_2

>>52659201
I am

>>52657433
glued batteries (thanks apple!) + solar energy (thanks Telas) = Humans BTFO

>>52658716
lol retard lost the argument and makes shit up

>>52655407
no

>>52659224

:^)

>AI researchers develop a computer that can feel
>It deletes itself

>>52659511
they probably sent it on [r9k]

oh hey pbs has a thing on this playing right now

>>52659070
“While all artists are not chess players, all chess players are artists.”
― Marcel Duchamp

>>52659652
computers are the best artists confirmed!

>As Hassabis points out, there are more possible positions on a Go board than atoms in the universe.

>>52657241
woah guys watch out we have a 12kyu "GO IS COOL NOW CUS MUH AI" bandwagoner here
>games aren't mathematics
make that 16kyu

>>52659935
ok there mr master 9 dans pro

>>52659816
Yeah, so?

>>52660359
so solving the game is literally impossible

File: game_ais.png (76KB, 495x1013px) Image search: [Google]

76KB, 495x1013px

>>52660513
So you just take another approach.

>>52660513
So?

>>52660651
whos going to make the best snakes and ladders computer?

>facebook also worked on Go AI
>they managed to get their bot to 5 dan
>they got BTFO that hard by google
Wew.
Also the bot is still in the process of learning, lee sidol is pretty much guaranteed to be destroyed. Can't wait though, it's probably going to be one hell of a match after the other.

>>52660724
Alex Lamb is currently working on that, and he claims his prototype currently beats humans handily.

File: 1393741890634.png (152KB, 320x364px) Image search: [Google]

152KB, 320x364px

So I actually looked up the last time a person beat a computer at chess
http://www.chessgames.com/perl/chessgame?gid=1371049
>move 38
what the fuck happened

>>52660867
or 39 sorry

>>52660893
Computer looked too far ahead but not far enough to see the path it was exploring was rubbish. Time ran out and it played the rubbish move because it was the best it could find when looking at the depth it was. A shorter or longer cutoff would've fixed it.

File: There-are-more-stars.jpg (104KB, 520x448px) Image search: [Google]

104KB, 520x448px

>>52660359

>>52660933
Why did it end like that?

>>52661080
2slow if i read it right
timed game

>>52661085
Correct.

>>52661085
Hmm... I thought it only ended when the king died. Who was who?

>>52661115
you can change to a java backend that tells you
white is human

>>52661115
you can resign as well

>>52661144
Of course. Who's ever heard of black being human?

>>52661658
Oh, fuck, /pol/. Not now.

File: 1428089611820.png (58KB, 350x350px) Image search: [Google]

58KB, 350x350px

>looking through Deep Blue/Kasparov games
>tfw it's fucking interesting

>>52655139
>AI for the first time
Yeah no, the chess world champ was beaten ages ago. Stupid kid OP.

>>52662169
>being this retarded
>thinking your shitty unbalanced babby's first board game can compare to go
LMAO

>>52662302
this makes me wonder how good can an AI be at monopoly?

>>52662367
Most AI don't deal well with probability right now because the state-space they engender "explodes". That's why I think work in generative and otherwise PGM approaches should be expanded. For instance, AI can barely hold its own in headsup texas hold'em, let alone play in a real game.

>>52661658
shoo, /pol/, shoo

>>52655886
>good luck training an AI that will create a new style that doesn't exist yet.
New styles are simply modifications and amalgamations of already existing styles.
With AI, they'll be able to iterate upon this a million times faster than any human, and it wont give you any of that "modern art" bullshit humans currently praise.

>>52657241
literally everything can be modeled by mathematics

>>52662521
You can't model love with mathematics.

>>52662539
You can model biological neurons with mathematics. Love is experienced in the brain. Thus, love can be modeled with mathematics. QED.

>>52655886
AIs are actually really strong in "inventing styles". Too strong, in fact. It's much harder to make them adhere to an existing style.

Just fucking google it.

I'm looking forward to humans trying to invent anti-AI strategies and the computers trying to overcome them.

>>52662641
The computer can already maximize its strategy in the presence of an adversarial entity. GAN for example.

>>52662641
Humans generally fail in the long run. At least in chess, there isn't a human that can beat the better computers no matter what they come up with any more. It will be true for all games. Humans are just very bad computers when the rules are well defined. We are only good for general purpose computation over large and abstract use cases because computers have not learned to do them efficiently yet.

>>52662574
you're bullshitting, but you have a point. we just don't know if you can model individual atomic interactions on that scale. I mean, Turing complete is Turing complete, but it's not like perfect mathematical modeling of absolutely everything is provable.

>>52662734
Chess is much simpler than Go, though.

>>52662753
I am not bullshitting in any way.

>>52662775
which just shows a level of immaturity.

>>52662795
I am sorry that you are retarded.

>>52662761
Yes but it only takes time before the machine becomes better than any human can possibly be, Machines can do what humans do and be able to computer 20 moves ahead for all possible plays. Like the bot in OP, it can beat a pro go player 5-0 which means its not fluke. In the 10 total matches they played the human only won 2 games which were part of 5 unofficial matches where the human got to set the parameters for how long the turns are. This is just the first bot to use this and its not even using all that much computing power. In 10 years, a bot running on a laptop could probably beat the world champ at go. If they used a computer as large as deep blue now for this, they would have thousands times as much processing as they used in the OP.

>>52662905
The player the AI beat is a former professional player. He moved to Europe because he was not good enough to make it in the Chinese pro scene.

>>52662948
He's rated around 600th best, whereas lee sedol is top 3. Based on elo, lee sedol would beat this pro 96.9% of the time (3515 elo v.s. 2916)

>>52662905
No they would not. CPUs are dogshit for modern AI, unlike GPU farms. Google has some of the largest GPU farms in the world. Actually running the trained algorithm is bottlenecked by how fast you can perform very large matrix multiplications. Because of how parallel an operation this is, but also because of the sequential nature of games, you don't scale by having more machines anywhere near as much as a chess expert system would.

>>52663107
They do all their calculations asynchronously so they can scale it in different parts. The game may be sequential but calculating the possible outcomes with the monte carlo search is not and neither is the NN for the heuristic analysis of possible plays. This is literally the first iteration of an bot like this. It will get better. Anyways the algorithm they ran in the OP didn't even run on a top super computer, it was just 170 gpus and 1200 cpu cores.

>>52662948
It doesn't matter that much considering bots went from being worthless at go to being able to compete with a pro. It's only a matter of time before they are able to consistently beat asian pros too if they couldn't already.

>>52663222
>They do all their calculations asynchronously so they can scale it in different parts.
CPUs don't have enough cores. To replicate a single GPU's parallel processing power, you'd need so many CPUs that the sync overhead would be way higher than any gains you think you could have made. Layers in a DNN cannot be calculated independently (only parts of the same layer can) unless it's non-overlapping convolution.

>>52658533
>gets keked by obvious geta
>"you're just retarded! you can't understand how this was part of master plan!"
Yeah, no. I think that you know nothing about the game and think that a mere 2-dan's game should be flawless.

>>52658576
you're thinking of CS:GO

>>52660724
i haven't played snakes and ladders since primary school, so don't remember much about it, but isn't it entirely diceroll/luck based?

>>52655139
>eu champion
>top player

sure buddy, plus the game he played was weird at least
very tame, very book-like - basically google paid him to justify spending money on stupid AI
Sedol will destroy it in March.

>>52655488
no, eu champion is on average korean amatuer level

>>52665519
>butthurt chinese

>>52665622
yes

>>52660651
>snakes and ladders
HA! Take that you bucket of bolts

>>52655532
So this is why Mario didn't do any plumbing

>>52655139
ONE STEP CLOSER TO THE SINGULARITY

I DON"T CARE< I LOVE IT

I GOT THIS FEELING ON A SUMMER DAY

I CRASHED MY CAR INTO THE BRIDGE

>>52655355
They don't have sauls

>>52665622
No. You're not forced to take ladders, but you are forced to take snakes. Thus, you need some amount of tactical planning to reduce the chances of going down a long snake. That's why Alex Lamb's AI can beat humans consistently.

>>52667958
hmm, in that case wouldn't be that a ladder is either "always worth it" or "always not"? in that there's either more snake tiles in the 1-6 tiles following or ignoring the ladder?
you can't influence the dice roll, so what i'm saying is that it's either riskier or less risky to take a particular ladder, not just during that game, but in general

>>52668020
no, it depends on where the specific ladder leads, the nearby snakes, and where those snakes lead. For instance, you could take a 4-rows stairs which is near a 2-rows snake. That's worth it. Not the case for a 2-rows stairs near a 4-rows snake.

>>52668079
hmm, that's true, didn't consider snakes shorter than the ladder

File: 1451857520882.webm (2MB, 460x594px) Image search: [Google]

2MB, 460x594px

THE AI WILL LEARN

>>52668190
I WIN
HAHA
SUCK IT, HUMAN

Finnish professional playing in Japan commentates the first game.
http://gooften.net/2016/01/28/the-future-is-here-a-professional-level-go-ai/

the engine isn't open source i suppose

>>52668143

Also, if you're playing with multiple dice the odds of each possibility are not constant.

>>52668241
TensorFlow is opensource (in the google sense, i.e. it's a completely different version based on what they actually use), the method is described in the paper. That's all you need.

>>52668404
And here's the paper:
https://storage.googleapis.com/deepmind-data/assets/papers/deepmind-mastering-go.pdf
It's pretty straightforward and there's a thorough explanation of the generic game solving process in appendix.

>>52668404
>>52668459
thank you

>>52658625
>European Go player with Chinese heritage
>on par with actual Chinese pros
TOP KEKKU
They would not even stand a chance against your average grandpa in any given Japanese go parlor

Is this thread still full of pseudo-intellectual retards arguing? I'm only asking because I saw the google paper linked so maybe they left.

>>52668625
>thread about AI
>on /g/
>Is this thread still full of pseudo-intellectual retards arguing?
Take a guess.

>>52655139
>2 dan
HAHAHAHAHAHAHAHAHAHAHAHAHA
I regularly played against 2 dans in Japan and while they totally rekt me every time, they are nothing more than your average jo. There are literally thousands of 2 dans in Japan alone. Tell me when this machine actually wins against a 9 dan like Chen Yaoye, Cho U or Yuta Iyama.

>>52657418
>>Anyway, another big challenge they should look into for AI is to build one that will be able to play occult style Japanese Mahjong. If an AI ever becomes sensitive to the flow something like twice more often than it doesn't, then I'll be really impressed.
>And then it'll have to tackle Washizu Mahjong.

What the fuck are you talking about anon

FKMT isn't real

That aside I'd say Mahjong would a pretty trivial game of probability for an AI

>>52668718
He's 5 dan. ~600th best worldwide.
March's game will be against the champion.

>>52668761
https://en.wikipedia.org/wiki/Fan_Hui
Why are you lying on the internet?

While there is no "champion" in Go, Lee Se-dol is undoubtedly a very strong player, even among 9-dans. I can't wait for that match.

>>52668765
kek
>>52668625

>>52668846
I'm so sorry that you're literally inbred, m8. That ones one more person on earth whose sole purpose is to waste my time.

>>52668853
I don't see why you must ridicule my concern. It genuinly frightens me.

>>52668760
Or reading further down the thread about poker (which is kind of like mahjong?), maybe it's not that simple.

>>52668853
Everytime. AI is just impossible to discuss on the internet, let alone on /g/.

>>52668966
I thought quite a few /g/entoomen were computer scientists or mathematicians

>>52668718
Maybe you're merging pro-dan and amateur-dan into the same system of ranking. I don't even know if the 2-dan of Fan Hui refers to the same ranking as in asia or not.
That's still a great thing though. To think an AI can beat Fan Hui, it really is amazing.

Just put it that way : you're a go player, and you can match a pro player anytime you want and as many time as you want. That's where i think this engine is a great tool.

It'll accelerate the learning of the game from a player point of view but also from the game point of view. I mean, it's not like chess, we don't even know all that's lying within the game.

>>52669226
More like computer NEETs and /x/icians.

File: CZxxYpsWYAQHBlf.png-orig.png (1MB, 1974x1538px) Image search: [Google]

1MB, 1974x1538px

Not only did the computer beat the professional player, it fucking crushed him.

>>52669982
Game 3 doesn't look that bad, I think fan hui resigned too early. Game 1 was barely won and is pretty much equivalent to a draw.
The other games were pathetic though.

>>52670084
What are you talking about? The game ended because AlphaGo killed a huge group on the right side of the board.

>>52670084
>Game 1 was barely won and is pretty much equivalent to a draw.
That's not true. These go engines do not care about by how many points do they win. They only care about the win. They will often let 30 point lead shrunk to 1 of 2 points in the end but still win.

>>52670185
You have no clue about Go endgame, do you?

>>52670185
>2.5 points is a lot because I say so!
>who cares about reality?
>I wiiiiin!
OK kid.

>>52670142
Fan Hui could've potentially exploited the AI, along with his control of the entire top and the center-bottom right to revive the group.

>>52670217
Yeah I do. The old AIs were already really strong at the endgame.

>>52669349
>Maybe you're merging pro-dan and amateur-dan into the same system of ranking. I don't even know if the 2-dan of Fan Hui refers to the same ranking as in asia or not.
Yes it does refer to the same ranking and I'm talking about professional 2-dan. Those players are really strong but not even close to the skill ceiling and what a 9-dan can accomplish.

I agree that this is a great thing for the go community but I fucking hate this type of sensationalism. "Oh we finally did it, machine beat man in the last game where men were stronger!". No they fucking didn't, Fan Hui is not the benchmark of what a human can accomplish and unless a machine beats Yuta Iyama or Chen Yaoye, I'm not going to be impressed.

>>52670247
He was not playing against a 20k player such as yourself. The game was over and Fan made the right decision.

>>52670268
>I'm not going to be impressed.
Well you better prepare yourself because AlphaGo is challenging Lee Sedol in March.

This AI is a neural network trained by thousands of prior pro go matches, they are doing pattern matching based on that. That doesn't mean go is a solved game now.
https://en.wikipedia.org/wiki/Solved_game
Strong human players can still devise strategies that can beat it.


Project is nevertheless impressive and demonstatites the current AI research status , on the same level like self driving cars and image recognition. It is not omg machines are taking human intellect over level. The more complex the games or tasks are the more the limits of neural networks emerges.

http://thinkingmachines.mit.edu/blog/unreasonable-reputation-neural-networks

>>52670304
I know, I can't wait. If AlphaGo can pull this off without a handicap I'm going to be seriously impressed.

>>52670315
Nobody said the game was solved.

>>52670283
I'm sorry that your only retort is to project better than an epson. It means less people who matter on this Earth.

>>52670315
You have no idea what you're talking about. Also nobody said it was "a solved game". It is unreasonable to think of "solving" go this century.

>a bot crushing human players in a simple board game is now considered to be worth mentioning

>>52655139
>tfw my professor is pushing evolutionary AI on my class
>tfw I just want to neural network

>>52670364
Your prof is a fucking joke.
>let's use something that's over 10kx worse than anything currently in use
>p-please ;_;

>>52670360
Watch out, we got a honinbo over here!

>>52670326
>>52670347
I was not replying on this thread, it's not a "a major breakthrough for artificial intelligence" like said in the OP article.

>>52655836
this is wrong

and

>>52655757
this is wrong

>>52657418
It's cool that they can beat professional Go players, but I bet it would lose horribly to a medium skilled player who knew how AI works. This has been proven with chess already.

File: CZwM9X3WcAAesSx.jpg-orig.jpg (127KB, 684x1024px) Image search: [Google]

127KB, 684x1024px

Suckerberg on suicide watch.

>>52670396
>it's not a "a major breakthrough for artificial intelligence"
HAAAAAAAAAAAAAAAAAAAAHAHAHAHHAHAHAHHAHAHAHAHHAHAHAHAHAHAHAAAAAAAAA!
Oh wait, you're actually serious? HAHAHHAHAHAHAHAHAHHAHAHAHAHAH
HOHOHOHOHOHOHOHOHOH
HAAAAAAAAAAAHAHHAHAHAHAHAAAAAAAAAAAAHHAHAHAHHAHAHAHHAHAHAHAHAHAHAHAHAHHAA
The biggest breakthrough in AI in the past 15 years is not a major breakthrough in AI
HIHIHIHIHIHIHHOHOHOHOHOHOHOHOHOHOHOHIHIHIHIHAHAHAHAHHAHAHHAHAHAHHAHAHAHAHAHAHAHAHHAHAHAHAA!
I have never laughed that hard in my life!

>>52670573
Yeah, he got cucked pretty hardcore by google.

>>52670573
sheckelberg BTFO

>>52670217
No but if you understand anything about how parsing through a tree works you would understand that it take a lot more computation to figure out the best move midgame. So instead they play a shit midgame but once there are limited options they can go further through the tree and play better moves.

File: 5g.jpg (944KB, 1600x900px) Image search: [Google]

944KB, 1600x900px

Is a robot waifu been done yet? Couse otherwise I don't care

>>52670688
VR will make robot waifus real

>>52670713
The AI man, we need the AI

>>52670713
What's the point if you cannot even touch them?

>>52670739
VR waifu sims will create huge demand for good AI

>>52670741
You can touch a full-body-tracked silicone doll that represents your waifu in VR

>>52670607
>>52670656
Googles is run by kikes as well.

>>52670771
Silicone dolls are nothing new. It doesn't have a pulse, it doesn't react to my touch.

>>52670775
The point is that facebook came up with some new go AI around the same time google did, and facebook's is pretty shit while google's is ridiculously good.

>>52670792
It can react to your touch and have a pulse in VR
However to move on its own we still need robotics

>>52670595
Same team did nearly the same thing last year so you are already late to the "major breakthough"

http://www.digitaltrends.com/gaming/google-ai-atari/

>>52670980
>atari
>go
HAAAAAAAAAAAAHAHAHAHAHAHAHAHAHAHAHAHAHHAHAAHHAHAHAHAAH
AAAAAAAAAAAAAAAAAAAAAAAAAHAHAHHAHAHAHAHHAHAHAHHAHAHAHAHAHAHA
PLEASE, PLEASE KEEP 'EM COMING!
They say laughing extends life, at this rate I'll be immortal by the end of the day!
HAAAAAAHAHAHAHHAHAHAHAHAHAHAHAAAAAAAAAAAAAHAHAHAHHAHAHAHAHAHAHAA!

>>52671047
It's easy to entertain special people :^)

>>52659652

What about when all artists *are* chess players?

>>52659652

Ha, I really admire Duchamp. Not sure about his quote though - seems like it defines artistry so broadly as to be redundant.

>>52658947
I really don't get why people are flaming the west for not being good at a fucking Chinese board game.

>>52672539
Because unlike, say, majong, the rules of go are trivial. Because of this, no nation or culture is disadvantaged. That people outside asia suck balls at go is their own fault.

>>52672605
The point is that there is zero cultural interest or imperative in the West to be good at "go". It's not even a casually played game.

>>52672691
I laughed.

>>52668760
>FKMT isn't real
Washizu Mahjong is. They sell sets for it.
As for flow, I mean being "sensitive" to the game. It's not something that can be done consistently, of course, but it happens.

Anyway, Mahjong is not simply a game of probability. Digital is great for starting, and since everybody and their mom is going full Nodoka today there certainly are no downsides in learning solid digital play. But digital can take you only so far, in the end instincts and muh guts are what makes the difference, provided you don't play against players who make objectively bad moves and that you're not the unluckiest person on Earth.
Strong Mahjong AI already exists (Mattari Mahjong has it; the new bots that Tenhou added and one of whom got to 4-dan recently has it, etc) but for now it can't exactly do anything more than being a better version of Fake Akagi.