How the World Got Wired · Why Hinton Went North

The Long Education

Geoffrey Hinton was a strange student.

He started at Cambridge University as an undergraduate in 1967, intending to study physics and chemistry. He hated it. He switched to architecture. He hated that too. He switched to philosophy, then to psychology, then to physics again, then back to psychology. He finally graduated, in 1970, with a degree in experimental psychology — a field that, in the late 1960s, was beginning to be transformed by people thinking about the human mind in computational terms.

He worked, for a year, as a carpenter. Then he went to graduate school in Edinburgh, in Scotland, where he completed a PhD in artificial intelligence in 1978.

At a time when most of his classmates were studying expert systems — programs that worked by following long lists of explicit rules written by human experts — Hinton was already interested in the unfashionable alternative: neural networks. The approach that Minsky and Papert’s book had supposedly killed in 1969. The approach that almost nobody in serious academic computer science was working on.

After his PhD, Hinton did a postdoctoral fellowship at the University of California in San Diego, where he joined the Parallel Distributed Processing group you read about in the last chapter. Rumelhart, McClelland, Hinton, and several others spent years there working out the mathematics of how multi-layer neural networks could be made to learn.

In 1982, Hinton took a faculty position at Carnegie Mellon University in Pittsburgh, Pennsylvania. He stayed for five years. During those years, he worked on the backpropagation paper that was published in Nature in 1986 — the paper you read about in the last chapter, the one that quietly listed the Office of Naval Research in its acknowledgements.

It was while he was at Carnegie Mellon, with his most important scientific work either complete or close to it, that Hinton started thinking seriously about leaving.

The Decision

What follows is a sensitive part of the story, and I want to tell it carefully, because Geoffrey Hinton has told it himself, in interviews and public talks, over the past several decades, and what I am about to describe is in his own words, not my speculation.

In the 1980s, the United States was deep into what was then called the Cold War — the long political and military rivalry between the United States and the Soviet Union. President Ronald Reagan, elected in 1980, had launched an enormous buildup of American military spending. A great deal of basic scientific research, at major American universities, was funded directly or indirectly by the Department of Defense — by ARPA (which was, at that time, called DARPA, with the D for Defense), by the Air Force, and by the Office of Naval Research you read about in the last chapter.

This was not, by itself, unusual. The American government had been funding scientific research through military channels since the end of the Second World War. Many scientists took the money without much thought. Some scientists took it with mixed feelings. A few refused.

Hinton, by his own account, became increasingly uncomfortable. He has said, in many places, that he did not want his research on neural networks to be used to build weapons. He has pointed out that the same techniques that could make a computer recognize faces in friendly contexts could, in different hands, be used to recognize faces in military contexts — for surveillance, for targeting, for things he did not want to be part of. He has said he found the political atmosphere of Reagan-era America uncomfortable, and the dominance of military funding in his field troubling.

He has also said — this is important — that he was not making any general claim about everyone else’s choices. Other scientists at Carnegie Mellon took the same Defense Department money and did fine work with it. Hinton was not telling them they were wrong. He was making a personal decision about what he could live with.

In 1987, he resigned from Carnegie Mellon and moved to Canada.

The job he took was at the University of Toronto, where he became a professor of computer science. He would remain at the University of Toronto, in some capacity, for the next thirty-five years.

The Canadian government funded his work through an agency called the Natural Sciences and Engineering Research Council, or NSERC. The grants from NSERC were modest. They came in the form of small annual stipends — tens of thousands of dollars per year, not the much larger contracts that American researchers in fashionable fields could attract. But the money came without conditions about military applications. It came from the broader Canadian public — taxpayers in a country that, while still part of NATO, had a different relationship with its military than the United States did.

Hinton has said, simply, that he could work on what he wanted, without his work being aimed at weapons. That was the point.

All 24 chapters of How the World Got Wired · Library · All courses