DART: the scheduler that repaid thirty years
A transport planner built in ten weeks during Operation Desert Shield was in use at US Transportation Command and US European Command through the winter of 1990–91. It reworked plans for moving troops and cargo at the speed at which those plans were changing.
Why it matters
The loudest number in the history of AI funding comes from military logistics: by the estimate DARPA's then director gave a reporter, DART repaid within a few months everything DARPA had put into artificial intelligence over the previous thirty years. What ended the AI winter was not an argument about intelligence but a transport schedule.
DART began as the idea of a BBN Systems and Technologies programme manager as Desert Shield got under way, and was built with Ascent Technology, SRA and ISX in a ten-week effort that compressed what had been planned as eighteen months, under a DARPA contract with US Transportation Command as sponsor. BBN's final functional description of December 1992 places the intensive development in the autumn of 1990, driven by needs the command identified during Desert Shield, with the concept of operations published on 30 October 1990 and the functional specification on 10 November 1990, and the product in use at USTRANSCOM and USEUCOM during the winter of 1990–91. It was the first demonstration system of the ARPA-Rome planning and scheduling initiative formed in 1990 with the Air Force Rome Laboratory. The repayment claim needs care. It comes from a 2002 magazine article, which attributes to Victor Reis, DARPA's director at the time, the statement that DART paid back all of DARPA's thirty years of investment in AI in a matter of a few months. That is a retrospective estimate quoted in a trade magazine, not an audit; confidence is therefore medium and the figure is given as an estimate rather than a measurement.