DeepMind AlphaGo: AI Teaches Itself 'Thousands of Years of Human Knowledge' Without Help

DeepMind researchers have developed an artificial intelligence program that has taught itself “thousands of years of human knowledge” from scratch in just a few days.

The AlphaGo Zero algorithm, created by the Alphabet (formerly Google) subsidiary DeepMind and detailed in the journal Nature this week, mastered the classic strategy game Go without human input, while reaching a level never before reached by humans or computers.

In a blog post by DeepMind co-founder Demis Hassabis and AlphaGo researcher David Silver, the algorithm is described as “arguably the strongest Go player in history.”

Keep up with this story and more by subscribing now

DeepMind's AlphaGo algorithm was able to master Go—a game with more possible board configurations than there are atoms in the known universe. DeepMind

AlphaGo Zero was able to reach a high standard of the game by using a novel form of reinforcement learning, which effectively enabled the algorithm to teach itself.

The program begins by playing games against itself until it learns to predict moves and eventually the winner of the games. Each time it plays, the algorithm improves by a small amount until after 40 days it reached a virtually unbeatable level.

“Over the course of millions of AlphaGo vs. AlphaGo games, the system progressively learned the game of Go from scratch, accumulating thousands of years of human knowledge during a period of just a few days,” the blogpost reads.

“AlphaGo Zero also discovered new knowledge, developing unconventional strategies and creative new moves.”

Hassabis previously described DeepMind as the Apollo Program for artificial intelligence—referencing NASA’s efforts to put a man on the moon—with its primary mission being to “solve intelligence and use it to solve everything else.”

In January the AlphaGo algorithm became the first computer program to beat a human professional Go player. Google DeepMind

Hassabis says the latest breakthrough will ultimately have real-world applications and will help push scientific advancements.

“It’s amazing to see just how far AlphaGo has come in only two years,” said DeepMind co-founder Demis Hassabis. “AlphaGo Zero is now the strongest version of our program and shows how much progress we can make even with less computing power and zero use of human data.

“Ultimately we want to harness algorithmic breakthroughs like this to help solve all sorts of pressing real world problems like protein folding or designing new materials. If we can make the same progress on these problems that we have with AlphaGo, it has the potential to drive forward human understanding and positively impact all of our lives.”