PROSPECTOR was a computer-based consultation system for mineral exploration developed at the Artificial Intelligence Center of SRI International (then Stanford Research Institute). The team, which included Richard Duda, Peter Hart, Nils Nilsson, John Gaschnig, Rene Reboh, and others, described the system in SRI work dated October 1977 and in a published paper the following year. PROSPECTOR was designed to help exploration geologists evaluate the mineral potential of a site during the early, expensive stage of deciding where to drill.
The system worked by capturing the knowledge and reasoning of expert geologists in networks of inference rules. A geologist would describe the rocks, minerals, and geological setting observed at a site, and PROSPECTOR would combine that evidence through its inference network to estimate how well the site matched models of known types of ore deposits. To cope with the uncertainty inherent in geology, where evidence is partial and judgments are probabilistic, the system propagated certainties through the network rather than treating facts as simply true or false, an approach in the same spirit as the certainty factors used in MYCIN.
PROSPECTOR’s celebrated result was that it identified a previously unknown molybdenum deposit in Washington State, a concrete demonstration that an expert system could deliver a finding of real economic value. Alongside DENDRAL and MYCIN, it became one of the canonical proofs that domain expertise could be encoded and made to pay off.
Why a business reader should care: PROSPECTOR is an early case study in turning scarce human expertise into reusable software that supports high-stakes investment decisions, and its mixed legacy, a striking success against a backdrop of brittle, hard-to-maintain rule bases, foreshadows the realities of deploying knowledge-based AI.