Undercover Algorithm: A Secret Chapter in the Early History of Artificial Intelligence and Satellite Imagery

Abstract

In the second decade of the twenty-first century, computer algorithms accurately and rapidly identify features of some objects on digital satellite imagery. These feature-recognition algorithms are expected to transform geospatial intelligence: to enable rapid retrospective searching of imagery archives and focus prospective analytic attention. This Perspective article establishes the beginning of the U.S. Intelligence Community’s research and development on this capability. Archival research of declassified Central Intelligence Agency documents produced two discoveries: one identifies that the earliest feature-recognition initiative predated the creation of the National Photographic Interpretation Center (NPIC) in January 1961. The other discovery reveals that the earliest neural network software, Frank Rosenblatt’s Mark I Perceptron, from which current feature-detection software descends, had been part of a previously secret four-year NPIC effort from 1963 through 1966 to develop this algorithm into a useful tool for photo-interpreters. The manager of that research effort, John Cain, defined the prospective utility of this software in 1963, and Cain’s criteria, derived from NPIC’s experience during the Cuban Missile Crisis, continues to shape the prospective geospatial intelligence uses of feature-recognition software.