A probabilistic theory of trust concerning artificial intelligence: can intelligent robots trust humans?

Abstract

In this paper, I argue for a probabilistic theory of trust, and the plausibility of “trustworthy AI” in which we trust (as opposed to mere reliance). I show that the current trust theories cannot accommodate trust pertaining to AI, and I propose an alternative probabilistic theory, which accounts for the four major types of AI-related trust: an AI agent’s trust in another AI agent, a human agent’s trust in an AI agent, an AI agent’s trust in a human agent, and an AI agent’s trust in an object (including mental and complex objects). I draw a broadly neglected distinction between transitive and intransitive senses of trust, each of which calls for a distinctive semantical theory. Based on this distinction, I classify the current theories into the theories of trust and theories of trustworthiness, showing that the current theories fail to model some of the major types of AI-related trust; while the proposed conditional probabilistic theory of trust and theory of trustworthiness, unlike the current trust theories, is scalable, and they would also accommodate major types of trust in non-AI, including interpersonal trust, reciprocal trust, one-sided trust, as well as trust in objects—e.g., thoughts, theories, data, algorithms, systems, and institutions.