Lee Se-dol, the second-highest ranked Go player in the world, was confident he’d win. Then he lost. Badly.
The Ancient Greeks may not have had our technology, but that didn’t stop them from imagining tales of conscious, humanoid beings just like us. Hephaestus fashioned Talos, a bronze automaton, who was sent to protect Europa on Crete. Daedalus created living statues that were so life-like, they had to be tied down to prevent them from running away. Whichever the era we live in, the idea of machines that perfectly simulate, or even exceed humans, is a concept that’s deeply rooted in our psyche.
So imagine the excitement when AlphaGo, a program developed by Google-owned DeepMind, beat legendary Go (a game also known as ‘wei qi’) player Lee Se-dol 4-1 in a five-game series. The win ignited headlines across the globe, with news outlets trumpeting the outcome as a major milestone for artificial intelligence. In South Korea, coverage of the match was even more prominent than North Korean threats of a pre-emptive strike on the South.
AlphaGo’s victory is a big deal, and not just because it’s the first computer program to triumph over a top-ranked Go player. By proving that machines can rival humans in an intuition-based game like Go, AlphaGo just overcame a huge hurdle for artificial intelligence (AI), a milestone that some thought was a decade away.
To be sure, AlphaGo is good at one thing only, playing Go, and is a long way off from being considered an artificial superintelligence. But if AI wasn’t expected to be so successful so early, what other surprise developments lay in wait? Progress doesn’t follow a linear curve, and every so often, it leaps forward exponentially. Is AlphaGo’s victory a sign that artificial superintelligences are closer to being realized than previously thought?
"ALPHAGO’S VICTORY IS A BIG DEAL, AND NOT JUST BECAUSE IT’S THE FIRST COMPUTER PROGRAM TO TRIUMPH OVER A TOP-RANKED GO PLAYER.”
WHAT IS GO?
To the casual observer, AlphaGo’s triumph over one of the world’s leading Go players may not seem like much. It’s just a game, right? But the win is significant in many ways. That’s because Go is a game unlike any other, where the rules are sparingly simple, but there’s also room for a mind-boggling amount of complexity.
A game like chess can be overcome with simple heuristics; decision-making shortcuts that involve ranking possible moves and picking the best one, based on elements like the material value of each piece.
With Go, however, the board provides limited data, and each piece has the same value as the next (discounting all other factors). Instead, Go requires a great deal of intuition to master, a quality that one would not usually attribute to computers.
Furthermore, the sheer intricacy of Go, where there are more possible Go positions than atoms in the universe, means that it is literally impossible for computers to just brute force their way to a win.
That makes AlphaGo’s win different from other victories by machines over humans, such as when IBM’s Deep Blue beat chess grandmaster Garry Kasparov back in 1997. In fact, Go’s reliance on intuition means that masters of the game cannot even fully articulate what makes them so good. In the words of Michael Polanyi, a Hungarian-British polymath,“We know more than we can tell.”