How the World Got Wired · A Naval Office in Arlington

The Boom

The first great wave of artificial intelligence research happened in the 1950s and 1960s.

In 1956, a young computer scientist named John McCarthy convened a workshop at Dartmouth College in New Hampshire to study what he called artificial intelligence — a new phrase he had coined for the proposal. Several brilliant young people attended. They believed, with some justification, that by understanding how the human mind worked and replicating it in computers, they could build machines that would think within a generation.

For about fifteen years, the field was wildly ambitious. Researchers built programs that could play chess, prove geometric theorems, solve simple puzzles, parse simple sentences. Each new result was greeted as a step toward true machine intelligence. Funding poured in — from ARPA, from ONR, from elsewhere.

In 1958, a psychologist at Cornell University named Frank Rosenblatt built a particular kind of machine called a perceptron. The perceptron was, in physical form, a closet-sized rack of electronics. It was the first device that could learn to do a simple task — recognizing patterns in pictures — by adjusting itself through experience, rather than being programmed step by step. Rosenblatt demonstrated it to journalists. The New York Times, in 1958, reported breathlessly that the Navy had revealed the embryo of an electronic computer that would be able to walk, talk, see, write, reproduce itself and be conscious of its existence.

This was, of course, overstating things considerably. The perceptron was a simple early version of what we now call a neural network — a kind of computer program inspired by the way networks of neurons in the brain work. Rosenblatt’s machine could learn to tell, for example, the shape of a triangle from the shape of a square. It could not learn anything much more complicated. But the idea was new and important: a machine that could be taught through examples, rather than programmed through rules.

For about a decade, perceptrons and other neural networks were a hot area of research.

Then, in 1969, the field collapsed.

The Winter

Two researchers at MIT — Marvin Minsky and Seymour Papert — published a book called Perceptrons. The book was, in part, a careful mathematical analysis of what Rosenblatt’s machines could and could not do. It pointed out, with proofs, that single-layer perceptrons — the simple kind Rosenblatt had built — could not learn certain extremely basic patterns. Some of the simplest things a child could learn to recognize, the perceptrons mathematically could not.

The book was technically correct. But it was, in a sense that you should understand, devastating in ways that went beyond its mathematics. Minsky and Papert were the two most famous AI researchers in the world at the time. They had decided that perceptrons — and, by implication, the entire neural-network approach — were a dead end. Their book was widely read as a verdict, even though they had not fully explored more complicated, multi-layer versions of the same idea.

After Perceptrons, almost nobody worked on neural networks for the next decade. Funding for the approach dried up. Graduate students were told it would ruin their careers. The promising researchers from the perceptron era moved on to other things. Frank Rosenblatt died in a boating accident in 1971, on his 43rd birthday, before the field had a chance to recover.

This is what happens, sometimes, in science. A whole approach to a problem gets discredited, often unfairly, by a single influential book or paper. For a decade, or two decades, almost nobody works on it. Other approaches get tried. Most of them also fail. The field as a whole goes quiet, and money goes elsewhere.

The 1970s, in artificial intelligence, were one of these quiet periods. Researchers called it, later, the first AI winter. Government funding for AI dropped sharply. The grand promises of the 1960s had not been kept. The machines that were supposed to think had not learned to think. Disappointed funding agencies turned to other things.

But not all of them. ONR, and a few other patient institutions, kept funding small numbers of researchers who had not given up. The amount of money was much smaller than it had been. The researchers were less famous, less ambitious in their public claims. They were doing what scientists do during a long winter: working quietly, on hard problems, without much hope of immediate success, because they thought the problems were worth working on.

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