Pilot's Associate: a minute of flight, two hours of computing

In March 1988 the Lockheed and US Air Force team, and in June the McDonnell Aircraft team, gave the second demonstration of the Pilot's Associate, a DARPA and Air Force programme begun in February 1986 for Strategic Computing: expert systems assisting a pilot in a simulated cockpit. At McDonnell fully integrated software supported only one of six mission segments: a minute of flight took about two hours of processing.

Teams led by Lockheed-Georgia Company and McDonnell Aircraft Company began development after a competition in fiscal 1986, per DARPA's budget of February 1987. The budget of February 1986 lists the start of the competitive procurements for the project among the accomplishments of fiscal 1985 and says in its programme for fiscal 1986 that an architecture of cooperating expert systems is being designed for it. A year earlier the programme's First Annual Report (February 1985) lists under Pilot's Associate only two preparatory awards: a functional mission definition (Perceptronics) and a laboratory definition study (Draper Labs). The Second Annual Report (February 1986), reprinted in a book and read in snippets, adds that specifications for the first development phase went to industry in early 1985 and that a competitive procurement chose two teams (selected in February 1986, per Roland and Shiman) that share in the cost of the project; the snippets read give no sums. Lockheed gave Demo 2 in a laboratory cockpit driven by six Symbolics machines and a VAX and, since it was not in real time, produced a real-time replay of the mission. Lizza's paper of October 1989 calls Lockheed's Demo 2 a successful demonstration of the fifth prototype, seen by over 200 pilots, scientists and managers from DoD and NASA. By Banks and Lizza's IEEE Expert paper (June 1991) the Demo 2 system took about 180 minutes for the 30-minute mission, about one sixth of real-time speed, that is six times slower. A paper for a workshop of September 1990, written by two Air Force officers and two staff of Search Technology, a subcontractor of the Lockheed team, puts the Lockheed system at about six times slower than real time in both Demo 2 and Demo 3. McDonnell worked on six Explorers and a microVAX in ART from Inference Corporation. Its repeat, Demonstration 2X, was set for February 1989 with a goal of 10 to 1: a single-ship mission against a column of tanks at a nominal 200 feet and 450 knots, the aircraft modelled on the F-15, the cockpit in a 40-foot dome. Lockheed held Demo 3 in November 1989; McDonnell's was not scheduled until October 1990. After Demo 2 the programme was redirected to a current or planned service aircraft and to near-term hardware; by the AGARD paper and by Roland and Shiman it was DARPA that gave the order, while Banks and Lizza say only that the programme was redirected. For that reason Lockheed's Demo 3, first set for the third year, moved to the fourth. By Banks and Lizza it was a functional prototype that ran in nonreal time and still lacked the knowledge depth for full functioning; two separate real-time demonstrations were shown with it, but the live system itself stayed nonreal-time. LaPuma and Marlin (1993) sum up Phase 1 as two laboratory demonstrations in non-real time on available hardware. Roland and Shiman write that at Demo 2 the McDonnell team made a poor showing and the Lockheed team performed spectacularly; they date it as a whole to March 1988, the Lockheed date, while the McDonnell team's own paper of 1991 puts its Demonstration 2 in June 1988. Contract values: no primary document read gives one, and the figures of later works differ. Roland and Shiman write that Lockheed was awarded $13.2 million and McDonnell $8.8 million, and that both companies contributed 50 percent of the cost. A commentary in The Cyborg Handbook (1995) gives $4.2 million to McDonnell and Texas Instruments and $6.7 million to the Lockheed team for Phase 1 in 1986, with no source visible; the snippets read do not say whether either pair is the full value or DARPA's share. A Lockheed engineer spoke in 1995 of a $43 million programme between 1986 and 1992, and a Canadian review of 2008, without a source, of the programme as part of a $40 million Lockheed effort in the same years. The record adopts none of these sums. What the record does not claim: that later demonstrations truly ran in real time (the authors of the AGARD paper write that the 2X system was a functional skeleton and that claims of real time must be judged with that caveat); that any of the sums above is a contract value. Banks and Lizza's IEEE Expert paper was read only in snippets of the Internet Archive's full-text search, with printed page numbers (IEEE's text is paywalled, the Archive's copy is lent only and was not borrowed); the books of Roland and Shiman (their sentence on the sums split by a page break and read in two snippets) and The Cyborg Handbook likewise; the papers of 1989 (Lizza), 1991 (the McDonnell team) and 1993 (LaPuma and Marlin) only on their free first pages. The account of the McDonnell team rests on the Wright Laboratory paper for AGARD and the first page of the team's own paper. The programme's annual reports after February 1986 were not found on any open host. No document read names a demonstration in 1987: Roland and Shiman write that in the fall of 1987 both teams were preparing the first one for the following March.