The Fifth Generation project ends

The Fifth Generation Computer Systems project held its final conference in Tokyo in June 1992 and released its software into the public domain. It had built a series of parallel inference machines, the last of which could perform some hundreds of millions of logical inferences a second, and a parallel logic-programming language to run on them. It had not built a computer that understood speech, translated languages or reasoned about the world, which is what the 1981 announcement had promised and the West had feared.

The New York Times called it Japan's lost generation. The judgement was harsh and, in the terms the project had set for itself, fair. The machines were obsolete on delivery because ordinary microprocessors had improved faster than the specialised hardware, a pattern that had already killed the Lisp machine and would recur with every attempt to build a chip for one kind of AI.

The end of the project closed the era of national programmes for symbolic intelligence. The next government money at that scale went into the internet, and when governments returned to AI in the 2020s the thing they were funding was the opposite of what ICOT had tried: statistical, learned, and running on hardware built for graphics.