Data Science Investment Counter —Funding raised by UK data science companies in 2018.
£ 5.640 Million

DeepMind’s New AI Doesn’t Need Humans to Learn Stuff

 

5700107240_88fe707d7f_b

Go stones, as those AlphaGo Zero will use to humiliate a human player

DeepMind has announced that the latest version of its Go-playing artificial intelligence has learnt to excel at the board game without help from humans.

Alphabet-owned (but London-based) AI lab DeepMind hit the headlines last year when its programme AlphaGo defeated top-class Go player Lee Sedol; since then , the AI has repeatedly triumphed over some of the planet’s best players.

The AI’s neural networks had gained mastery of the ancient Chinese game over several months, by sifting through  thousands of games played by humans. This has now changed radically.

DeepMind’s new creation, AlphaGo Zero, acquired its Go skills simply by analysing the game’s rules, and then playing repeatedly against itself. It took only three days for it to grow skilful enough to trounce the original AlphaGo 100 games to zero.

What is more, the machine appeared to be coming up with completely new playing strategies, instead of building on techniques previously observed in human champions— as AlphaGo did.

“The new version starts from a neural network that knows nothing at all about the game of Go. The only knowledge it has is the rules of the game. Apart from that, it figures everything out just by playing games against itself,” explained AlphaGo lead researcher David Silver.

“[AlpahGo Zero’s success] shows it’s the novel algorithms that count, not the compute power or the data.”

He added that, if the technique were successfully applied to other fields—he suggested science and medicine— future smart machines could be able to assimilate thousands of years’s worth of human knowledge and skills in a matter of days.

“We’ve actually removed the constraints of human knowledge and it’s able, therefore, to create knowledge itself from first principles, from a blank slate,” Silver said.

 

Image via bit.ly/2yyEdqa

SHARE THIS ARTICLE:
BY SHACK15

Co-working space and blog dedicated to all things data science.

Subscribe to our newsletter