Post 3: The movement 37 of AlphaGo 🧐
Published:
In 2016, DeepMind (creators of AlphaFold2) created “AlphaGo,” an AI that learned to play the board game Go. DeepMind put it to compete in 5 games against the best player of the decade (named Lee). In one of them, AlphaGo made a famous move known as “The move 37,” which went against intuition and even seemed like a bad idea, but that move paved the way for the AI to win the game several rounds later. In fact, the AI beat the human in 4 out of the 5 games.
Move 37 caused Lee to be caught off guard, and he even left the room to think and smoke a cigarette. Even the analysts were caught off guard when they saw the move (the full documentary of this event is on YouTube).
Training AIs to learn to play board games was the first step. Today we have AlphaFold2. We even have another AI called “Gato” that knows how to play board games and about 600 other things.
And all that, in less than 6 years 😬
Since AlphaGo surpassed human skills, people have improved their decision-making and learned new game strategies. That is, a super AI specific to a topic (Go) allowed people who already knew how to play well to play even better.
- Superhuman artificial intelligence can improve human decision-making by increasing novelty
- Open access paper
What will happen with AlphaFold2 and protein science? 🤔 The DeepMind team has a blog with several “success stories” of what has been achieved thanks to AF2.