Precursors
In 1981 the Japanese Ministry of International Trade and Industry announced the Fifth Generation Computer Systems project. The idea was to leap over the existing architecture of computing entirely — past the von Neumann machine, past conventional programming — and build machines designed from the silicon upwards to perform logical inference, to reason, and to be programmed in a language of logic rather than of instructions. It launched in April 1982, run through a new institute called ICOT in Tokyo, staffed with young researchers seconded from Fujitsu, Hitachi, NEC, Toshiba and the rest, on a ten-year plan.
The budget was somewhere between fifty-four and fifty-seven billion yen, which at the exchange rates of the day was on the order of three hundred and twenty to four hundred million dollars over a decade. I want to dwell on that figure, because it is the most instructive thing about the whole episode. It was less than one per cent of the annual research spending of the Japanese electronics and communications industry. IBM alone spent more on research every year than the entire Fifth Generation project spent in ten. It was, financially, a rounding error.
Rhetorically it was an earthquake. The American response was immediate and disproportionate. The Microelectronics and Computer Technology Corporation was formed in Austin in late 1982, explicitly as a defensive move, under the direction of a former head of the National Security Agency; such a consortium was actually illegal under American antitrust law at the time, and Congress passed the National Cooperative Research Act in 1984 substantially in order to permit it. The Defense Advanced Research Projects Agency launched the Strategic Computing Initiative in 1983, which over a decade spent around a billion dollars of defence money in pursuit of, in the programme's own language, a machine that would run ten billion instructions per second and see, hear, speak and think like a human being. Britain launched the Alvey Programme, about three hundred and fifty million pounds. Europe launched ESPRIT. Every one of these was a direct response to Tokyo.
And the Fifth Generation project produced almost nothing of commercial consequence. It built some genuinely interesting parallel inference machines, a concurrent logic programming language, an operating system, a theorem prover, several hundred patents and a couple of thousand technical reports. It trained a thousand engineers, which was probably worth the money on its own. But its architectures were commercial failures, general-purpose microprocessors overtook special-purpose symbolic hardware while the project was still running, and the bet on logic programming cut it off from the graphical interfaces, networking and eventually the internet that actually defined the following decade. It wound down between 1992 and 1994 and the software was given away.
The American programmes did rather better, though not at what they set out to do. The Strategic Computing Initiative's machine-intelligence goal failed by the late 1980s; what it produced instead was real progress in autonomous navigation, and a logistics planning tool that was credited during the 1991 Gulf War with savings that the agency has claimed repaid thirty years of its artificial intelligence funding in a single deployment. The consortium in Austin spun out fourteen companies and dissolved in 2000.
The moral is not that government programmes fail, which is too easy and also false. The moral is about the ratio. A modest, badly-aimed research programme in one country produced, in three others, a decade of industrial policy, a change to American antitrust law, a billion dollars of defence spending, and a genre of book about impending economic supremacy. The response was two orders of magnitude larger than the stimulus. And the entire episode has vanished so completely from memory that people now say, quite regularly and in print, that the current moment is the first time governments have treated artificial intelligence as a matter of national economic survival.
