DeepMind’s AI Can Now Beat Humans at More Board Games


More than a year after DeepMind introduced AlphaZero—a self-taught, world champion algorithm that can kick your butt at chess, shogi, and Go—the AI firm is celebrating another milestone.

CEO Demis Hassabis and a team of engineers on Monday announceda the full evaluation of AlphaZero, published in the journal Science.

The short paper (only five pages) describes how the latest evolution of Google’s elite algorithm quickly learns each game to become the strongest player in history, despite starting from scratch each time.

“AlphaZero replaces the handcrafted knowledge and domain-specific augmentations used in traditional game-playing programs with deep neural networks, a general-purpose reinforcement learning algorithm, and a general-purpose tree search algorithm,” according to the document.

In other words, AlphaZero has no need for humans.

The platform learns by playing against itself—with no intervention or historical data. And it’s a quick study.

Starting with a blank slate, the neural network knows nothing about the game, aside from basic rules. But through regular, self-contained gameplay, the system incrementally improves, creating stronger versions of itself.

“This ability to learn each game afresh, unconstrained by the norms of human play, results in a distinctive, unorthodox, yet creative and dynamic playing style,” the DeepMind team wrote in a blog post.

Traditional chess and shogi (Japanese chess) engines rely on thousands of man-made rules. AlphaZero, meanwhile, takes the opposite approach, replacing those guidelines with unschooled algorithms.

“It’s like discovering the secret notebooks of some great player from the past,” said Chess Grandmaster Matthew Sadler, who, along with Women’s International Master Natasha Regan, analyzed thousands of AlphaZero’s chess games for the upcoming book “Game Changer.”

In each evaluation, AlphaZero convincingly beat its opponent (via DeepMind)

Developed by Alphabet, Inc.’s Deep Mind, the software in 2015 became the first computer Go program to beat a human professional. It made history again a year later, when it pummeled Lee Sedol, and triumphed once more in 2017 with a three-game win against then-No. 1 ranking player Ke Jie.

The latest iteration, however, differs from its predecessors: AlphaZero abandons all hand-engineered features, runs only one neural network (versus the two found in earlier models), and relies solely on its own knowledge to evaluate positions.

The amount of training the network needs depends on the style and complexity of the game, DeepMind said. While chess and shogi can take between nine and 12 hours, respectively, Go requires as many as 13 days.

“Chess has been used as a Rosetta Stone of both human and machine cognition for over a century,” former World Chess Champion Garry Kasparov said in a statement. “AlphaZero renews the remarkable connection between an ancient board game and cutting-edge science by doing something extraordinary.”

More on Geek.com:

Site Search 360 Custom Site Search

Be the first to comment

Leave a Reply

Your email address will not be published.


*