Knowledge

Knowledge is true belief held in a way that makes its truth no accident. The classical analysis spelled that out as justified true belief; Gettier’s counterexamples showed that justification and truth can coincide by luck, and the decades since have produced anti-luck conditions, virtue theories, and Williamson’s proposal that knowledge is not to be analysed at all but taken as basic.

Most of what anyone knows was learned from others. That makes the epistemology of testimony central rather than specialised, and it is where my own work sits. A patient asks a chatbot what dose of acetaminophen is safe for her ten-year-old. The chatbot answers correctly; she gives the dose; the child is fine. She came to a true belief, acted well on it, and relied on a source that was accurate. Did she learn from the chatbot as she would have learned from her paediatrician?

I argue that she did not, in that sense, and the answer has two parts of different strengths. The constitutive claim is that on the best current evidence these systems do not transmit testimonial knowledge, because the truth of what they assert does not figure, through their own operation, in why they assert it. That is an empirical claim, held with the confidence the evidence supports and no more. The practical claim is that hearers are not now positioned to treat model outputs as testimony, and it does not wait on the first.

This is not scepticism. The patient may well know the dose, if she used the system as a competent reader uses a well-validated instrument, and instruments are genuine sources of knowledge. What she cannot do is defer, as a hearer may to an informant whose warrant she inherits. Whatever knowledge she has, she generated by her use, and the load of getting it was on her (see instrument reading).