Making Trust Safe for AI? Non-agential Trust as a Conceptual Engineering Problem

Abstract Should we be worried that the concept of trust is increasingly used when we assess non-human agents and artefacts, say robots and AI systems? Whilst some authors have developed explanations of the concept of trust with a view to accounting for trust in AI systems and other non-agents, others have rejected the idea that we should extend trust in this way. The article advances this debate by bringing insights from conceptual engineering to bear on this issue. After setting up a target concept of trust in terms of four functional desiderata (trust-reliance distinction, explanatory strength, tracking affective responses, and accounting for distrust), I analyze how agential vs. non-agential accounts can satisfy these. A final section investigates how ‘non-ideal’ circumstances—that is, circumstances where the manifest and operative concept use diverge amongst concept users—affect our choice about which rendering of trust is to be preferred. I suggest that some prominent arguments against extending the language of trust to non-agents are not decisive and reflect on an important oversight in the current debate, namely a failure to address how narrower, agent-centred accounts curtail our ability to distrust non-agents.