Diagnosis and three demonstrations

What was missing

The atlas’s medical line began with MYCIN in 1974, although attempts at diagnosis by machine had begun fifteen years earlier, and between 1982 and 2007 the line had two records. The military line had the 1983 Strategic Computing plan, which promised three demonstrations: a land vehicle, a pilot’s associate and a battle management system. The atlas had none of them. The direction the 6-in-1 set is the same: the prehistory of a line matters more than its tail.

Nine records

Medicine has four records. Ledley and Lusted’s paper in Science on 3 July 1959 broke diagnosis into logic, Bayes’ formula and value theory. The Leeds study of 1972 tested machine diagnosis at the bedside on 304 patients with acute abdominal pain: the computer was right in 91.8% of cases, the most senior clinician in 79.6%. CASNET, a model of glaucoma, was rated by 49 ophthalmologists in 1976. PUFF, built on a shell taken from MYCIN, interpreted lung function tests daily at a San Francisco hospital from 1979.

Three records belong to Strategic Computing. The ALV land vehicle drove a road by itself in May 1985 and began steering around obstacles in October 1986. The FRESH expert system reached the Pacific Fleet headquarters in August 1986. The Pilot’s Associate’s second demonstration in 1988 ran about 120 times slower than real time: a minute of simulated flight took two hours of computing. Two more records are industrial: General Electric’s CATS-1 for repairing locomotives (1983) and the Authorizer’s Assistant, which decided on credit at American Express from 1988.

Where the document said less

As before, I wrote the hypotheses before opening any source. Three agents looked for the documents, and I checked every claim against a verbatim excerpt and the full text. Ledley and Lusted’s paper has no measurement at all: its numbers are illustrative, so the record is about how the problem was framed, not about accuracy. Warner’s 1961 paper could be read only as an abstract without figures, and it did not enter. For the ALV’s first demonstration two primary documents give different speeds, 5 and 3 km/h, and the record gives both. The Authorizer’s Assistant was described twice by the same author, and one paper dates its full deployment to 1988, the other to 1989. The hypotheses included the Leeds system’s 600 earlier patients, CASNET’s 70% agreement and real time for the Pilot’s Associate. None of these is in the documents.

Without a browser

The defence department’s site would not serve the DTIC reports, so they were read in the original scans in the Internet Archive. DARPA’s budget of February 1987 was scanned sideways and upside down, and its pages were read as images. The Internet Archive lends two IEEE journal issues only, so just the snippets of its full-text search were read, and the sources say so. The axis page’s budget was raised as the first action. After the package the tightest is the organisations page, with 694 bytes left.