Google DeepMind’s ‘superhuman’ AlphaZero computer taught ITSELF to beat humans at chess, shogi and Go

GOOGLE has created an artificial intelligence so clever it can teach itself to become a “master” of chess.

Normally AI learns based on existing data – like watching chess masters player – but AlphaZero earned “superhuman mastery” just by playing against itself.

Google DeepMind creates artificial intelligence software to tackle difficult problems – like puzzle games
AFP or licensors

The AI was created by DeepMind, a London-based company owned by Google – responsible for creating genius AI.

AI, or artificial intelligence, is a type of computer program that normally require human intelligence.

This can be anything from recognising speech, perceiving visuals, making informed decisions, and translating languages.

In a new study, DeepMind revealed how AlphaZero could master some of the “most complex board games” – like chess, shogi and Go.

AI has already proved adept at mastering chess – and even more difficult games like shogi and Go
EPA

Being able to beat humans at their own games is considered a benchmark for AI.

Often the best way to do this was through “machine learning”. This involved have AI watch endless games being played, and learning over time what strategies worked best.

But for AlphaZero, the AI was trained without “no prior knowledge” except for each game’s rules.

It suggests that game-playing AI can now learn to play and master any game.

Decades ago, IBM’s Deep Blue chess program beat the human world champion.

But since then, AI has grown significantly more advanced, capable of playing more abstract games like shogi and Go.

Both of these games are deemed to be “significantly more difficult” than chess, but have now been mastered by machines.

However, AI has typically relied on “handcrafted” knowledge from developers, rather than by teaching themselves.

Google’s DeepMind says that AlphaZero “forgoes the need for human-derived information”.

“It was able to learn chess, shogi and Go by playing against itself – repeatedly – until each was mastered,” the official report explains.

And although AlphaZero is focusing on games for now, the applications for the future go far beyond simply winning a game of chess.

”The hardest challenge in AI is building systems that are ‘general’ which means they can tackle more than one problem,” a Google DeepMind spokesperson told The Sun.

“While humans find this very easy – and can switch from driving a car to cooking a meal to writing an email without effort – AI programs don’t.

”AlphaZero is exciting because it represents a step on the journey to more general systems, which in the long run could help society unlock some of the hardest problems the world faces, from understanding diseases to reducing energy consumption.”

AI now sees these complicated games as “easily solvable problems”.

This means Google’s DeepMind team now needs to look to harder games for its AI to tackle.

An obvious challenge would be multiplayer games – like Call of Duty – which has an enormous number of variables, and would be difficult for an AI to master.

Do you think AI will inevitably take over the world? Let us know in the comments!


We pay for your stories! Do you have a story for The Sun Online news team? Email us at tips@the-sun.co.uk or call 0207 782 4368 . We pay for videos too. Click here to upload yours.


***

Note from WSOE.Org : This content has been auto-generated from a syndicated feed.