Reliabilism and the Testimony of Robots

Abstract

We are becoming increasingly dependent on robots and other forms of artificial intelligence for our beliefs. But how should the knowledge gained from the “say-so” of a robot be classified? Should it be understood as testimonial knowledge, similar to knowledge gained in conversation with another person? Or should it be understood as a form of instrument-based knowledge, such as that gained from a calculator or a sundial? There is more at stake here than terminology, for how we treat objects as sources of knowledge often has important social and legal consequences. In this paper, I argue that at least some robots are capable of testimony. I make my argument by exploring the differences between instruments and testifiers on a well-known account of knowledge: reliabilism. On this approach, I claim that the difference between instruments and testifiers as sources of knowledge is that only the latter are capable of deception. As some robots can be designed to deceive, so they too should be recognized as testimonial sources of knowledge.

AI-Testimony, Conversational AIs and Our Anthropocentric Theory of Testimony

“…Additionally, they illustrate that, unlike humans, computers are not (yet) sensitive to small errors that humans will immediately notice. build upon Nickel (2013a) and Wheeler (2020) to introduce the notion of 'quasi-testimony', but do not further develop that notion. Taking a cue from their work, and to continue research on human perceiving natural language outputs by technology and specifically conversational AIs, I suggest the concept of 'AI-testimony'.…”

Over What Range Should Reliabilists Measure Reliability?

“…The basic format for the cases I discuss here will be that of a user acquiring beliefs based on a machine learning system. 1 As Wheeler ( 2020 ) argues, these cases can be understood as testimony that a user receives from the AI system. To really isolate this process of testimony, I consider only the case where the user is presented with very little information aside from the system output.…”