CONCEPT
Assessability
Onora O’Neill’s standard for trustworthy communication—not that speakers are sincere but that audiences are given adequate means to evaluate what they receive—and the specific condition that AI-generated output systematically fails to meet.
Sincerity and assessability are not the same thing, and the distinction is the hinge of Onora O’Neill’s most consequential contribution to thinking about trust and communication. Sincerity is an inner state: the speaker believes what she says. Assessability is a structural property of communication: the audience has adequate means to evaluate the claims being made. Sincerity is morally relevant but epistemically invisible—you cannot tell from outside whether a speaker believes her own words. Assessability is observable and therefore actionable: a communication is assessable when the speaker provides evidence for her claims, identifies her sources, acknowledges uncertainty where uncertainty exists, and avoids rhetorical devices that make weak claims appear stronger than they are. These practices do not guarantee truth. They guarantee something more practically useful: they give the audience the information necessary to place their trust on the basis of evidence rather than on the basis of surface presentation. The arrival of
large language models has made assessability the most urgent concept in applied epistemology. AI