Digital duplicates are AI systems that imitate real, living or dead, individuals. Digital duplicates are currently being produced and marketed for entertainment and educational purposes, and they have been proposed for medical care and scientific research, as well. They are commonly described as simulations of their subject’s “mind,” “personality,” or “communication style,” but these targets of simulation can come apart. Scientists regularly use simulation models that represent real systems in distorted and simplified ways, but because such models misrepresent their target systems, scientists are expected to justify their reliance on them. I argue that digital duplicates warrant similar scrutiny.

This paper examines digital duplicates in light of the debate, in the philosophy of science, on the epistemology of simulation models, and argues that research on digital duplicates ought to pay closer attention to their specific targets of simulation; that developers have a burden of proof to establish the adequacy of digital duplicates for the purposes for which they are marketed; and that digital duplicates (if they are developed) should be developed so as to discourage their application to purposes for which they are not adequate.