The Student Room Group

A robot is the best chess player in the world after four hours of teaching itself

http://www.bbc.co.uk/news/technology-42251535

AlphaZero, an AI engine created by DeepMind in London was given the rules of chess and played itself for four hours. After that it played 100 games against a world leading chess engine Stockfish 8 and won or drew every match (28/100 of the games were wins).

DeepMind AI engines have previously become masters of Space Invaders, Pong and the boardgame Go by training and improving over time. AlphaZero has been called "superhuman" because it was trained without the use of datasets derived from human expertise. AI is no longer constrained by the limits of human knowledge and the future possibilities are endless/scary.
Reply 1
A win for the machines! I want to see AI like EDI from Mass Effect, that would be great.
Absolutely terrifying:afraid:

:hide:

This is the only robot I will tolerate in society:
I thought Kasparov beat all these robots
(edited 6 years ago)
Original post by Notnek
http://www.bbc.co.uk/news/technology-42251535

AlphaZero, an AI engine created by DeepMind in London was given the rules of chess and played itself for four hours. After that it played 100 games against a world leading chess engine Stockfish 8 and won or drew every match (28/100 of the games were wins).

DeepMind AI engines have previously become masters of Space Invaders, Pong and the boardgame Go by training and improving over time. AlphaZero has been called "superhuman" because it was trained without the use of datasets derived from human expertise. AI is no longer constrained by the limits of human knowledge and the future possibilities are endless/scary.

They had to program to program itself? I wld argue preprogrammed..otherwise computers will take over slow human thinker nd rule all o us
Reply 5
There are people who dedicate their entire lives to learning chess and now there are robots who can wipe the floor with them with only 4hrs practice...surely they must find that just a little depressing? I think we need to integrate robotics into humans ASAP!
Reply 6
...Now imagine they teach a robot warefare and how to operate weaponry etc O_O
Reply 7
Original post by Hirsty97
I thought Kasparov beat all these robots

You're thinking of 1997 when Kasparov played against Deep Blue but actually Deep Blue won that series (3.5 - 2.5), which was a key event in the development of AI.

Since then AI has got better and better and humans are no match. The reason why this story is significant is that an AI program managed to teach itself chess by starting from zero basically and playing against itself millions of times for four hours until it was better than all the other AI.
(edited 6 years ago)
Reply 8
Original post by Flying birds
They had to program to program itself? I wld argue preprogrammed..otherwise computers will take over slow human thinker nd rule all o us

Yes the AI was given the rules of chess and loads of other clever stuff. But it got better by playing itself and wasn't limited by human knowledge.
Original post by Notnek
http://www.bbc.co.uk/news/technology-42251535

AlphaZero, an AI engine created by DeepMind in London was given the rules of chess and played itself for four hours. After that it played 100 games against a world leading chess engine Stockfish 8 and won or drew every match (28/100 of the games were wins).

DeepMind AI engines have previously become masters of Space Invaders, Pong and the boardgame Go by training and improving over time. AlphaZero has been called "superhuman" because it was trained without the use of datasets derived from human expertise. AI is no longer constrained by the limits of human knowledge and the future possibilities are endless/scary.


If anybody could beat this robot, the chances are they would probably be the sort of person who has no friends or any form of social life. :sadnod:
Original post by Synchronise
If anybody could beat this robot, the chances are they would probably be the sort of person who has no friends or any form of social life. :sadnod:


My dad went to school with the dude that started Deep Mind, and said he was a chess geek. Now that he has hundreds of millions of dollars after selling Deep Mind to Google, I think it's a fair bet that he doesn't need friends - he just needs people to fish all the champagne corks and condoms out of his infinity pool.
a remarkable achievement.

today chess, tomorrow .... who knows ?

https://www.thesun.co.uk/wp-content/uploads/2016/04/1699082.main_image.jpg?strip=all
I think this is pretty amazing, personally - imagine the advances we could have if we applied the same upscaling to healthcare etc
I see the zealots and "journalists" that've watched too much sci fi and barely know how to operate a computer have caught wind of deep learning.
Of course they think neural networks are some kind of living creatures.
Original post by Notnek
http://www.bbc.co.uk/news/technology-42251535

AlphaZero, an AI engine created by DeepMind in London was given the rules of chess and played itself for four hours. After that it played 100 games against a world leading chess engine Stockfish 8 and won or drew every match (28/100 of the games were wins).

DeepMind AI engines have previously become masters of Space Invaders, Pong and the boardgame Go by training and improving over time. AlphaZero has been called "superhuman" because it was trained without the use of datasets derived from human expertise. AI is no longer constrained by the limits of human knowledge and the future possibilities are endless/scary.


This is indeed impressive. So I typed "canal" into my Google Photos. It did indeed pick several pictures of canals, but also picked pictures of rivers and bizarrely a picture of a miniature train.

Chess may have many combinations of moves, but at any point in the game, the number of move options are very limited. This isn't true AI. Just a neat algorithm.

Quick Reply

Latest

Trending

Trending