Fuck AlphaGo, the Koreans made a hive mind of robot arms that learns(SKYNET). It's over, humanity is done and finished.
https://youtu.be/l8zKZLqkfII
http://nextbigfuture.com/2016/03/deep-learning-for-robots-robotic-arms.html?m=1
>>7922240
>drop ship warehouses can now get rid of employees who have to find x product in x seconds to get it shipped
>>7922240
>fuck alphago
>implying this won't be used to make a robotic go player for sale at go clubs that places its own stones and removes its own captures
>>7922240
>It's over, humanity is done and finished.
kind of glad desu
we don't deserve this universe
>>7922240
Peter Norvig already proved strong AI impossible in this book. Don't go around spouting bullshit memes until you've read norvig in its entirity.
You are now aware that the true destiny of the human race was simply to bring about the birth of godlike AI.
>>7922784
There is a 0% chance you have read that book.
>>7922791
>he's a bayesian
I think I've seen this episode of Lexx.
>>7922791
of course he didn't, did you see how fucking tiny it is?
>>7922791
elaborate, please
Daily reminder MCTS and convolutional neural nets are trivial mathematics and any math major could have accomplished this thing which took CS majors decades over a weekend if math majors could just be bothered to solve such trivial problems.
>>7922784
>saving the thumbnail
im pretty sure alpha go is more intelligent than you ever will be.
>>7922839
>we could have done this thing that we didn't do
every time
>>7922899
watson beat chess players back in the 90s, before that it was known that you could optimize a strategy for a given set of rules in a game (particularly useful with Correspondence chess back when that was a thing)
This is more of a show of raw computing power than anything else. Much more interesting AI developments are in the realm of getting them to learn things and recognize patterns. But even then, it's still just taking input, reducing it to a number, and running it as a math problem. It's not really thinking about it, since it's not obtaining thought.
>>7922795
>he isn't a Bayesian
>>7922789
I hope it'll allows us to piggy-back a ride towards eternity.
>>7923082
>IT DOESN'T FEEL LOVE
Who the fuck cares, the goal is making it work for us.
>>7922774
Bullshit
we deserve to travel the stars
>theres an OLD world plan trying to stop humanity from doing everything it is to be human
>>7923113
lol being this deluded. if you think you are rebelling the jew then you need to rebuild your beliefs from the ground up
>>7922784
>proved strong AI impossible
>published 1994
http://www.syracuse.com/news/index.ssf/2014/11/technology_history_internet_computers_phones_1994.html
2016
http://www.ibtimes.co.uk/googles-ai-neural-network-robot-so-smart-it-has-solved-puzzle-used-google-job-interviews-1544729
It just takes time and information.
>>7923082
It is learning though.
You're right that when it's playing it is just running the opponent's moves through an algorithm and it probably has relatively beefy hardware to do so, that's not the impressive part.
Even if you think that you could develop an algorithm to be good at Go (good luck considering there are [math] 10^{741} [/math] possible games of Go), that's not even what they did. They developed an algorithm that could teach itself how to play Go by playing against itself over the last five months. When it made decisions that resulted favorably, it reinforced those decisions. When it made decisions that didn't result favorably, it weakened them. This is similar to how the human brain learns things.
And if you think that there were no mathematicians who worked on DeepMind then you're intentionally being difficult.
Why systems like that not be tested on other popular games like poker?
Deep learning fag here, this is neither profound nor scary.
>>7923381
Well they do
https://en.wikipedia.org/wiki/Computer_poker_players#Artificial_Intelligence
But poker is a lot more difficult to collect data on I'd imagine. Training on just the moves themselves would be pretty unfair for the computer, you'd need full motion capture of body language and facial expressions since that's something human players regularly make use of.
AlphaGo will always win. The problem isn’t the combination of moves: That is like saying you are impossible because of the possible combinations of atoms in your body is more than the amount of atoms in your body to the amount of atoms in your body power – it is irrelevant to evolution. AlphaGo doesn’t work that way.
The problem is the intersection of the combination of moves from two players on to a feasible region of winning boards, and humans and computers learn functionally the same way. That means the training set is a much smaller set of possible moves that depend on the moves of actual games that have been played or can be played to a feasible conclusion. A computer excels at that simply because it can play more games.
Humans have no advantage given them by learning other things in the world. The closed world of the game gives an advantage to winning the game to the computer. The magical thinking reverence to intuitive players and other spooky stuff is just bullshit. As long as the world stays consistent, engineering beats intuition every time.
In addition, the entire game can be reduced to a few a priori rules (constraints on moves) that end in feasible games; it is just a matter of figuring out the combination of these rules your opponent uses – a training set of rule combinations. This a computer also does better.
Lastly, AlphaGo doesn’t stop learning. Once AlphaGo beats you once, it is not likely to lose to you.
>>7922240
>smart robot arms
Yea, sure they can move around and pick things up but the real question is can they play Go?
>>7923420
OP here I had to hype it up. This does amaze me though. The fact that the robot seems to have learned multistep grasping techniques is amazing to me.
Even if they are using a fuckload of robots and are just applying known deep learning techniques.
I am skeptical how well this will generalize though. It probably can't generalize to different robots and this approach probably won't work that well when the grasper is occluded.
>>7923425
I don't think it really help. Anyway I mean AI based on math poker stuff.
>>7922240
> Koreans
"The research on robotic hand-eye coordination and grasping was conducted by Sergey Levine, Peter Pastor, Alex Krizhevsky, and Deirdre Quillen, with special thanks to colleagues at Google Research and X who've contributed their expertise and time to this research. An early preprint is available on arXiv."
http://googleresearch.blogspot.lt/2016/03/deep-learning-for-robots-learning-from.html
>>7923082
Do you realize how big the search space for Go is compared to Chess? It is barely at all about increasing computer power.
>>7923909
Well fuck, I thought it was kaist. I wish it was kaist.
Dammit I don't want cloud robotics!
>>7925341
Do you realize how big the search space for robot motion planning is? It's continous!
>those dam choppers hands at it again
why haven't we designed something even remotely close to our own hands?
>>7925399
because the control scheme is insane and our linear actuators fucking suck. we are nowhere near developing a synthetic analog to human muscle.
>>7922695
https://www.youtube.com/watch?v=quWFjS3Ci7A
>>7925388
Almost as big as yo momma yo
