1981

Japan’s Fifth Generation Computer Systems

Japan’s Ministry of International Trade and Industry (MITI)’s Fifth Generation Computer Systems (FGCS) project was a targeted 10-year research investment in logic programming, parallel computing, and artificial intelligence. The $450 million targeted investment of government funds was designed to push Japanese computer manufacturers and researchers to the forefront of computing by subsidizing the development of technologies that were not commercially feasible at the start of the project but were thought to be game-changing after development.

Key to MITI’s plan was the development of data-flow and logic programming techniques that were not popular in the West but that many academics thought would be the next big thing. Such technologies were widely believed to be better suited to parallel processing, an approach for speeding computers by performing multiple operations simultaneously. Wrapped up in this approach was a huge investment in AI. MITI’s leaders wished to create computers that could speak and understand natural human language, represent knowledge, prove mathematical theorems, and reason as people do. MITI picked the technologies and the companies for the project. The software would be written in Prolog, an arcane computer language that had elegant mathematical properties but ran quite slowly.

Many US companies were spooked. After all, in the early 1980s, strategic investments by the Japanese government helped Japan to capture 70 percent of the world’s market for 64-kibibyte (KiB) dynamic RAM chips. Was Japan poised to repeat the experience, this time with AI?

Ten years later, MITI shuttered FGCS and published to the internet the software it had created. It turned out that the millions that Japan had invested to make Prolog run fast was no match for the billions that the US market had available to make Intel’s x86 architecture run even faster. Meanwhile, the very premise of using Prolog as the basis of AI was called into question. Summing up Prolog’s failure to deliver, AI researcher Carl Hewitt wrote: “The laws of thought are inconsistent.”

This parallel inference engine could run a single Prolog program simultaneously on 512 parallel processors.