Articulating the interplay between collective memory and artificial intelligence, this paper examines the SAILDART archive, a unique digital collection derived from the backup tapes of the Stanford Artificial Intelligence Laboratory (SAIL). Preserved somewhat serendipitously and not originally intended to serve as an archive, SAILDART has retained nearly the entire contents of the laboratory’s computers from 1972 to 1990. By exploring the archive’s origins, structure, selection and appraisal practices, the article sets the stage for a discussion on the nature of machine-recorded and machine-produced (collective) memory. It not only aims to contribute to the historiography of early AI but also raises questions about interpreting such a machine archive. As a method of archival interpretation, I propose to employ modern AI algorithms, thereby making them a medium for understanding and interpreting their own history.