Mar 8, 2017
Robots are going to take our jobs and make us look like fools while doing it
By Jeremy Kahn
The other day I was talking to Demis Hassabis, co-founder and CEO of DeepMind, the artificial-intelligence research company based here in London that is part of Google parent Alphabet. DeepMind is best known for creating software that has beaten many of the world’s top-ranked players at Go, the ancient Asian strategy game. We were talking about rules of thumb. You know, those life hacks that condense years, decades or eons of human experience into some simple heuristic: the 80/20 rule, or “red sky at night, sailor’s delight.” There are tons of them, from almost every field. They aren’t meant to always hold true — but they are supposed to be right often enough to serve as practical guides.
This upending of received wisdom has, at least initially, left human players profoundly shaken.
Because Go is such an old game — it was invented more than 2,500 years ago — there are plenty of rules of thumb about the best ways to play. And one of the most interesting things about AlphaGo, the software DeepMind built, is how often it flagrantly violates these rules. For instance, it chooses to cede territory around the perimeter of the board in situations when humans, based on strategies developed over generations, do the opposite.
This upending of received wisdom has, at least initially, left human players profoundly shaken. “After humanity spent thousands of years improving our tactics, computers tell us that humans are completely wrong,” the reigning world champion, Ke Jie, wrote on the Chinese social media site Weibo after losing to DeepMind’s latest Go-playing AI. “I would go as far as to say not a single human has touched the edge of the truth of Go.”
And it isn’t just Go. This is happening with other games where AI — powered by neural networks and self-teaching algorithms — is beating top humans. The poker pros who lost to Carnegie Mellon University’s AI, called Libratus, in No Limit Texas Hold ’Em in February, marveled at the software’s unconventional betting strategies.
That generations of human wisdom, earned by hard experience, may be about to be wiped out is a fairly depressing thought.
We’re starting to see some real-world examples, too: In 2016, Google used DeepMind to reduce the amount of energy required to cool its massive data centers by 40 percent. Google engineers were stunned the software could achieve this savings. But Hassabis tells me that they were even more flummoxed by how the AI had done it. The rule of thumb was that the most energy efficient way to cool a building was to run as few systems as possible, maxing out each one before bringing additional units online. But the AI didn’t do this. Instead, it turned on almost all the cooling systems simultaneously, but ran them at lower power levels, balancing the heat load across almost all of them.
This same pattern is very likely to be repeated in every field AI touches. That generations of human wisdom, earned by hard experience, may be about to be wiped out is a fairly depressing thought. It’s not enough that robots are going to take our jobs, they’re going to make us look like fools while doing it. But Hassabis tells me that, on the contrary, he sees potential for AI to unleash a golden age of human creativity.
Take Go again. Hassabis says it’s now clear that the level of human play had reached a kind of plateau. Why? Because those learning the game falsely assumed that with more than 2,500 years of experience playing the game, every strategy had been tried and the rules of thumb accurately distilled this history. Top players, he says, weren’t likely to experiment with wild new strategies, because — even subconsciously — they wouldn’t want to risk losing games and prestige if that experimentation failed.
Then along comes AlphaGo and suddenly anything seems possible. This realization has spurred the world’s top human players into what Hassabis calls “super-creative mode,” with many of them now experimenting with strategies similar to those AlphaGo has used — and developing innovative new tactics of their own. “I think we are going to see huge innovation in Go, and I mean human Go,” Hassabis says.
The same thing will probably happen as AI gets deployed to other domains, he said. He’s most excited about what this may mean for science — especially fields such as drug discovery or materials science. But it probably holds true in everything from finance to sports. Instead of simply showing us the error of our ways, AI might also help guide humans to new heights of achievement.
This post was originally published in the Fully Charged newsletter. You can sign up here.
