FRESH: an expert system at the Pacific Fleet headquarters
In August 1986 Texas Instruments and btg delivered to the headquarters of the Commander-in-Chief, US Pacific Fleet, at Pearl Harbor the first prototype of FRESH, an expert system that tracks the status and employment of ships and helps plan their use. DARPA's budget of February 1987 says the prototype is installed and demonstrated at the headquarters along with a natural language interface, in nearly the words with which its budget of February 1986 had described the programme for fiscal 1986, before the prototype was delivered.
Why it matters
The battle management demonstration promised by the Strategic Computing plan reached a real fleet command centre rather than a laboratory. According to a Naval Postgraduate School thesis, by the late fall of 1987 the system was considered operational, though with limited capability, and the command centre estimated that the manpower and time these calculations needed fell to between a quarter and a twelfth of those of manual methods.
FRESH belongs to the Fleet Command Center Battle Management Project (FCCBMP). By DARPA's budget of February 1986, knowledge engineering with the headquarters staff began in fiscal 1985, when the computing hardware for the project's testbed was also installed at the headquarters; by the programme sections of both budgets, of 1986 and of 1987, work had started on CASES, which assesses force capabilities, the second of the project's five expert systems. The system tracks almost three hundred ships. It was built by rapid prototyping, with a knowledge engineer on site and a development group in Dallas. The programme's First Annual Report (February 1985) says that Texas Instruments will develop FRESH and that it is the first of the project's five expert systems. The natural language interface IRUS-86, by BBN's paper in the programme's workshop proceedings (May 1986), was delivered to the FRESH developers in January 1986 and installed in a test bed area of the fleet's command centre in April, with a demonstration set for June. Roland and Shiman (2002) tell the same story in a section that, except where noted, draws on DARPA's Third and Fourth Annual Reports and its plan of 1987: the test bed was completed and run up in March 1986; FRESH Prototype One was installed in it in June 1986 and demonstrated successfully to DARPA and the Navy two months later with the IRUS natural language interface, whose vocabulary then held 5,000 words; by 1987 the enhanced system did in ninety minutes what had taken the CINCPACFLT staff fifteen hours, and the Navy used it routinely to monitor the readiness of its ships in the Pacific. A programme paper of 1989 also has a demonstration in August 1986, with a snapshot database. They list the project's five expert systems as FRESH, CASES, CAMPSIM, OPGEN and STRATUS, and call FRESH and CASES fully operational by the late 1980s. Discrepancies: DARPA's budgets of 1986 and 1987 expand FRESH as Force Readiness Expert System, while the same agency's First Annual Report, the thesis and Roland and Shiman say Force Requirements; DARPA each time speaks of five expert systems in the project, as do Roland and Shiman, the thesis of four. The budgets of 1986 and 1987 put the same sentence on knowledge engineering begun with the headquarters staff among the accomplishments of fiscal 1985 and of fiscal 1986 respectively, and the sentence on the prototype installed and demonstrated with the natural language interface in the programme for fiscal 1986 and for fiscal 1987: the agency carried the wording forward a year, so neither budget dates these steps. The programme sections of both budgets also say that work on CASES has started, while Roland and Shiman put its start in August 1987, nearly a year behind schedule, and the amended budget of February 1988 still plans a CASES demonstration at the testbed for April 1988. What the record does not claim: the estimate of saved manpower and time as a measurement, since it is the command centre's estimate as the thesis reports it, not a test report; the ninety minutes against fifteen hours of Roland and Shiman come from a history that rests, by its note, mainly on annual reports not seen here. The thesis is the work of an officer who studied the system, not a programme document; of DARPA's own reports only the First Annual Report, written before the system was delivered, the budgets of 1986 and 1987 and the amended budget of February 1988, which only names the project among the major milestones of fiscal 1987, were read; no later annual report and no report of the Navy's on FRESH was found, though the Third (February 1987) and the Fourth (November 1988) lie behind Roland and Shiman's account.