This week in Seoul, in what has been dubbed the “ultimate challenge” for artificial intelligence, a man will face off against a computer in the ancient game of Go. If you’re reading this, you may be tempted to cheer for the man. But this is the rare battle of wits in which you should actually root for the robot.
AI is the next frontier in applying intelligent machines to the solution of human problems. Go is a revealing test because it has long bedeviled AI programs. Computer algorithms can’t easily replicate the intuition and creativity that top players bring to the game. And the immense number of potential positions in a given match means that brute-force calculation -- of the kind that computers rely on in games like chess – isn’t a practical strategy.
AlphaGo, a project of Google DeepMind, uses a different approach. It fuses two methods of artificial intelligence. One, called a deep neural network, helps the AI recognize patterns by imitating the structure of the human brain. The other, called reinforcement learning, helps it improve decision making through trial and error.
Combined, they present a powerful problem-solving tool. In a study published last year, DeepMind researchers described how an AI program learned to play dozens of old-school video games, beating humans in many of them. In a subsequent study, the group revealed that AlphaGo had beaten the European Go champion in October.
Now it will face Lee Se-dol, with 18 world titles to his name. He’s confident that he can win (“at least this time”), and it’s hard not to wish him well.
But an AI victory would be a stirring achievement. It would further validate the approach that DeepMind’s team is taking. It could provide substantial insight into the nature of human intelligence. And it would be a milestone for the problem-solving ability of computers, in a world with no shortage of problems to solve.
Google hopes its AI project will be a boon for health care, helping doctors and clinicians sort through the huge amount of data now at their fingertips. Financial firms, unsurprisingly, also find it alluring. It might one day work wonders in education, climate modeling, resource conservation and more.
The most thrilling possibility, though, is that AI will have uses no one’s thought of yet. Silicon Valley’s behemoths are investing huge sums in the technology. Many are offering open-source versions of their research and inventions, allowing developers to improve on them or come up with new uses for them. Start-ups in the field are proliferating. After decades of disappointment, the machines are starting to live up to the hype.
Is that something to fear rather than celebrate? There‘s no denying it presents challenges. Some of them are already familiar, such as adapting the economy to new patterns of work and ensuring that the gains from technological progress are widely shared. Others are still in the realm of science fiction, such as keeping humans in charge of their machines and not the other way around. Without making light of those concerns, one should also keep in mind the enormous potential that artificial intelligence -- a human enterprise, after all -- has for solving human problems.
That’s why a victory for AI this week wouldn’t be a defeat for humanity. It would be a thrilling triumph.
