Jobs. Driving. And machines that can learn
In the tech world, it was big news this spring that an artificial intelligence system won a game of Go against 18-time world champion Lee Sedol, triumphing in four out of five straight games of a five-game match. Aritificial intelligence — AI — had conquered chess in 1997, when Deep Blue, an IBM supercomputer, beat world champion Gary Kasparov. Go was the next grand challenge: A game exponentially more difficult than chess, and one that defied the brute-force methods in game tree search.
AlphaGo, the AI program that won the Go match, succeeded by doing things differently: It combined the usual tree search with search-space reduction techniques that use machine learning. In other words, it augmented brute-force search (computers’ traditional strength) with “intuition” that it developed by playing games against itself. Its victory is a stunning achievement — and a milestone in the inexorable march of AI research.
By relying on learned “intuition,” AlphaGo overcame an AI hurdle called Polanyi’s Paradox. In 1966, the philosopher Michael Polanyi observed, “We can know more than we can tell . ... The skill of a driver cannot be replaced by a thorough schooling in the theory of the motorcar.”
Some labor economists have viewed Polanyi’s Paradox as a major barrier for AI, arguing it implies a limit on its potential to automate human jobs. In 2004, economists argued that driving, in particular, was unlikely to be automated in the near future. But a year later, a Stanford autonomous vehicle won a DARPA Grand Challenge by driving over 100 miles along an unrehearsed desert trail. A decade later, both technology companies and car companies vigorously pursue the automation of driving. I expect the technical challenges to be resolved in the coming decade.
It is difficult for me to think of any computing technology other than automated driving that can be deployed in a decade or two with such benefit for humanity. About 1.25 million people worldwide die from car accidents every year. More than 90 percent of these accidents are caused by human error — by the sorts of inattentive or clumsy mistakes that a computer wouldn’t make. By automating driving, we could save over a million lives a year, as well as avoid countless injuries.
So far, we’ve approached automated driving largely as a technical issue, as if it carries no more societal weight than conquering the game of Go. But as AI matures, and as the technical questions are ever closer to being solved, it’s crucial that we consider ways to cushion the havoc this new technology will unleash on workers’ lives.
The solution to the labor problem will not be technical, but sociopolitical. We have a moral imperative to address it. Moshe Y. Vardi (@vardi), a Rice University professor, is director of the Ken Kennedy Institute for Information Technolog y. He is also editor-in-chief of the journal Communications of the ACM, where a version of this essay first appeared.