Dead End: Buying a Computer Industry (Unidata, the Fifth Generation, Alvey, and ESPRIT)

Abstract

Four times between 1972 and 1998, governments decided that a computer industry could be bought with a programme. Unidata (1973–1975) was three European firms told by their states to become one IBM, and it died when France changed its mind. Japan’s Fifth Generation Computer Systems project (1982–1994) spent just under ¥57 billion on parallel logic-programming machines that were obsolete before they were finished, and frightened the West into three responses: DARPA’s $1 billion Strategic Computing Initiative and the MCC consortium in the United States, Britain’s Alvey Programme, and the European Community’s ESPRIT, which alone outlived its cause and turned into the EU’s research framework. This article treats the four as one dead end with four exits, and asks what the programmes actually produced, which was usually not what they were for.

The Problem the Programmes Were Solving

By 1970 the System/360 had made IBM’s architecture the standard everyone else had to meet. Every other government with an electronics industry faced the same question: build a national champion, join forces across borders, or copy the leader. The Soviet bloc chose to copy (Comecon Computing). France chose a champion, CII, under the Plan Calcul of 1966 (France’s Tech Industry). Britain had already merged its firms into ICL (The UK Computing Industry). Italy had sold its division to General Electric (Italy’s Computing Industry). The next step, obvious on paper, was to put the champions together.

Unidata, 1973–1975

CII and Siemens agreed on 28 January 1972 to found a joint company, Philips joined by an agreement of 4 August 1972, and Unidata was founded in Amsterdam on 4 July 1973. The logic was complementary strengths: CII would bring software and architecture, Philips integrated circuits, Siemens peripherals and mechanical engineering, and together they would have the scale to build a common product line and take a European share of a market that IBM otherwise owned. It was the first attempt at a European champion rather than three national ones, and it lasted two and a half years.

What killed it was French politics. Ambroise Roux, chairman of the conglomerate CGE, wanted no new competitor and worked to see it disappear; Valéry Giscard d’Estaing, elected president in 1974, dismantled the government’s IT delegation on 2 October 1974 and then chose an American partner over a European one. On 20 May 1975 CII was merged with Honeywell-Bull, which took France out of Unidata, and the consortium was formally dissolved on 19 December 1975. Siemens and Philips went on alone; CII-Honeywell-Bull became Bull, kept alive by the state. The programme that was supposed to produce a European IBM produced a French subsidiary of an American one, and the lesson that a multinational champion can be killed by any one of its governments was learned in time for ESPRIT to be designed around it.

The Fifth Generation, 1982–1994

In 1982 Japan’s Ministry of International Trade and Industry launched the Fifth Generation Computer Systems project as a ten-year programme and set up the Institute for New Generation Computer Technology (ICOT) in Tokyo, with the Japanese computer makers as joint investors. The aim was a new kind of computer: massively parallel hardware running logic programming, measured in logical inferences per second, with a target of 100 million to a billion LIPS against the 100,000 or so a workstation of the day managed. Knowledge processing, natural language and reasoning would follow from the hardware. The software foundation was Prolog, reworked at ICOT into concurrent logic languages after Ehud Shapiro’s Concurrent Prolog of 1982 (see Prolog and Logic Programming).

The reaction abroad was larger than the project. Edward Feigenbaum and Pamela McCorduck’s The Fifth Generation (1983) presented it as Japan’s challenge to American leadership and sold the panic to Congress (see Expert Systems and the First AI Winter). In the decade that followed, ICOT built five Parallel Inference Machines, PIM/m, PIM/p, PIM/i, PIM/k and PIM/c, the Kappa database system and a body of concurrent logic programming research, and spent a little under ¥57 billion, about US$320 million, in all; at its peak year, 1991, the outlay was under one percent of the Japanese electronics industry’s own research spending. It ended in commercial failure. Ordinary microprocessors and workstations, improving on the curve Moore had described, overtook the special-purpose machines before they were finished; logic programming did not become the way software was written; the expert-systems market the project was meant to serve collapsed in the AI winter of the late 1980s. The project ended in 1994, and MITI and ICOT moved on to a neural-network programme that some called the Sixth Generation. The Japanese industry’s own account is in Japan’s Computing Industry.

The American Answer

The United States responded twice, once in public and once in private. In late 1982 a group of American computer and semiconductor makers founded the Microelectronics and Computer Technology Corporation (MCC), chose Austin, Texas, as its home in 1983 and put Admiral Bobby Ray Inman, lately of the NSA and CIA, in charge; the arrangement was illegal under antitrust law until Congress passed the National Cooperative Research Act in 1984. MCC did research in six areas from packaging to intelligent systems, became the largest customer of Symbolics and LMI during the Lisp-machine boom, and produced fourteen spin-offs and housed Doug Lenat’s Cyc from 1984; its board voted to dissolve it in June 2000. DARPA’s Strategic Computing Initiative, run from 1983 to 1993 under Robert Kahn’s Information Processing Techniques Office, put $1 billion into “machine intelligence”, ten billion instructions a second and machines that could see, hear, speak and think, with an autonomous land vehicle, a pilot’s associate and battle management as the showcase applications. By 1985 it had spent $100 million on 92 projects at 60 institutions. Jack Schwartz took over IPTO in 1987 and cut the AI funding sharply. What survived was, again, not the goal: the autonomous vehicle work is the ancestor of the driverless car, and the logistics planner DART, fielded in 1991, was credited with saving the Defense Department billions of dollars during the Gulf War.

Alvey, 1984–1990

Britain’s reaction went through a committee. A Department of Industry mission visited Japan in October 1981, and a committee under John Alvey, technology director of British Telecom, recommended that Britain emulate the Japanese programme rather than try to join it. The Alvey Programme that followed ran to 1990 under Brian Oakley as director, with government and industry funding collaborative research in four areas: VLSI and advanced microelectronics, software engineering, intelligent knowledge-based systems, and man-machine interaction. Alvey himself was not involved in running it. The programme funded university-industry projects across the country and trained a cohort of researchers, and when it ended there was no British successor; the collaborative funding moved to Brussels, and the largest British firm it had been meant to strengthen, ICL, was bought by Fujitsu in 1990. It is remembered in Britain chiefly as the moment the state briefly funded AI between the Lighthill cut of 1973 and the deep-learning revival.

ESPRIT, 1983–1998

The die of an Inmos T425 transputer; transputer-based parallel machines were among the projects ESPRIT funded. Image: Pauli Rautakorpi, CC BY 3.0, via Wikimedia Commons.

The European Community’s version was built to avoid Unidata’s fate. In late 1981 the industry commissioner Étienne Davignon invited the heads of the twelve largest European IT companies to round-table talks, and by early 1982 the twelve had a work plan for a joint research programme. After a pilot phase in 1983, the Council of Research Ministers approved ESPRIT in February 1984 with a budget of 1.5 billion ECU over four years, half from the Community and half from the participants. The design principle was pre-competitive research: companies from at least two member states would collaborate on technology below the level of products, so that no government could kill the programme by favouring its own champion, and no company had to give up its own product line. Five ESPRIT programmes ran consecutively from 1983 to 1998 and were succeeded by the Information Society Technologies programme in 1999; the Framework Programmes and today’s Horizon Europe are its descendants.

ESPRIT’s outputs were diffuse by design. Its projects included transputer-based parallel computers, the ProCoS and RAISE formal-methods work of the late 1980s and 1990s, the CGAL computational geometry library, the SUNDIAL speech-understanding project, the BBC Domesday project and a Web for Schools programme in the fourth phase. None of them produced a European IBM, or a European Intel or Microsoft, and it was not asked to. What ESPRIT did produce was a European research community that had worked together across borders, a template for collaborative funding that the Community kept, and, through Davignon’s round table, the political habit of treating information technology as a Community matter, which fed into the Single European Act of 1986. It is the one of the four programmes that is still running under another name.

Dead End: What the Money Bought

Set side by side, the four programmes fail in the same way and succeed in the same way. They fail at their stated purpose. Unidata did not become a European IBM; the Fifth Generation did not produce the reasoning computer; Alvey did not save the British computer industry; ESPRIT did not close the gap with the United States, which widened through the 1990s as the personal computer, the workstation and the internet were built almost entirely by American firms. The stated purpose was, in every case, to catch a leader by running a programme, and a programme cannot buy what makes a leader: a market of early customers, a labour pool that moves between firms, and a capital system that funds the second and third attempt (see Route 128 for the American region that discovered the same thing).

They succeed at something else. The Fifth Generation’s concurrent logic languages and a generation of Japanese parallel-computing researchers went into the supercomputers of the 1990s and 2000s; the SCI’s vehicle and planning work turned into products two decades later; ESPRIT built the European research area; even Unidata left the lesson that ESPRIT applied. The pattern is that industrial-policy programmes for computing return their investment through people and institutions and almost never through the product they were sold on. Governments that understood this, as the European Community did after 1984, kept funding. Governments that did not, as Britain did after Alvey, stopped, and got neither.