Book Chapter: 'Honest AI' in AI in Society: Relationships by N.G. Laskowski

In the forthcoming volume AI in Society: Relationships (Oxford University Press), MINT Affiliate N.G. Laskowski argues that conversational AI systems like ChatGPT, Gemini, or Claude should be designed not merely to be truthful, but honest. Drawing on moral philosophy and building from Thomas Hurka’s work on value asymmetries, the chapter develops a theory of AI honesty centered on vindicating human expectations.

Laskowski offers a detailed rebuttal of objections from Evans et al. (2021) and proposes a novel approach to alignment inspired by the ideal observer tradition in ethics. The result is a framework for designing AI that is not only safe and beneficial but normatively robust.

Read the chapter here.