Scientists create AI that can crush the world’s best AI (at board games, thankfully)


Nakhorn Yuangkratoke/EyeEm

Humans have mostly accepted that they will never be as good at chess as the robots, but now even the robots have to accept they will never be as good as other robots.

A new artificial intelligence platform, known as AlphaZero, can learn the games of Go, chess and shogi from scratch, without any human intervention. Using deep neural networks, AlphaZero quickly learnt each game “to become the strongest player in history.”

AlphaZero was unveiled by DeepMind Technologies in research published in Science on Nov. 6. DeepMind, a British AI subsidiary of Alphabet, Google’s parent company, has been tinkering with Go AI for a number of years. In 2017, DeepMind retired former AI champion AlphaGo, but continued tinkering with the AI. With AlphaZero, DeepMind’s research has reached its zenith.

The program was pitted against the world’s best AI for three board games: 

  • Stockfish, a world-champion chess AI
  • elmo, winner of the 27th annual World Computer Shogi Championship in 2017
  • AlphaGo Zero, DeepMind’s own Go AI touted as the strongest Go player in history

In each case, AlphaZero was only given the knowledge about the games basic rules. Before taking on the AI masters, it would then play millions of games against itself, starting off trying random tactics to win but slowly learning which strategies work best via a process known as trial and error called “reinforcement learning”. 

The training and learning process took nine hours for chess, 12 hours for shogi and 13 days for Go, involving 5,000 tensor processing units. For reference, just a single TPU can process over 100 million photos a day in Google Photos, so AlphaZero is a pretty heft piece of processing hardware. Once learning was complete, AlphaZero was unleashed on the AI competition.

az-blog-fig3-performance-width-1500

AlphaZero outperformed the world’s best AI in chess, shogi and Go.


DeepMind

And it crushed them.

What’s unique about the study is the fact that the learning algorithm was combined with a “searching method” called the Monte Carlo tree search (MCTS). This is a way that Go AI programs identify which move to make next. The DeepMind team used this same system for chess and shogi, showing for the first time that it could be adapted to other complex tested games.

Perhaps most intriguing for human chess players is the fact that AlphaZero, without human hands crafting its knowledge, implemented strategies and novel ideas that haven’t been seen before. Its aggressive style and highly dynamic play surprised chess Grandmaster Matthew Sadler, who spoke to the DeepMind blog.

Such unique strategies and abilities makes AlphaZero a great teaching tool for chess players — inviting hitherto-unseen tactical gameplay.

The AI-destroys-humans narrative is pretty consistent in the gaming world, with robots beating us at board games, complex multiplayer video games like Dota 2 and, of course, Go.

Does that mean AI are ready to beat us at literally every competitive game ever invented? Thankfully no. Although the three games utilised by DeepMind are remarkably complex, they provide some advantages for AI in that they involve two players and all the information necessary to make the next move is always visible.

So while they’ve definitely taken over as ancient board game champions, the robots likely won’t beat us at Texas Hold ‘Em.

(yet)

Techhnews’s Holiday Gift Guide: The place to find the best tech gifts for 2018.

‘Hello, humans’: Google’s Duplex could make Assistant the most lifelike AI yet.

Source link


Leave a Reply

Subscribe to our newsletter

Join our monthly newsletter and never miss out on new stories and promotions.
Techhnews will use the information you provide on this form to be in touch with you and to provide updates and marketing.

You can change your mind at any time by clicking the unsubscribe link in the footer of any email you receive from us, or by contacting us at newsletter@techhnews.com. We will treat your information with respect.

%d bloggers like this: