Conversation With an AI About Knowledge

3. Knowledge as generative competence: Instead of storing propositions with justifications, knowledge might be viewed as a capacity to generate apt responses. This resonates with enactivist and post-cognitivist accounts of cognition (e.g., Varela, Noë, Wittgenstein), where knowing is a kind of skill or practice.

4. Because LLMs are epistemically opaque, users must often rely on external indicators (e.g., accuracy over time, alignment with sources)

How It Should Affect Our Understanding:

1. Encourage a pluralist epistemology: Knowledge should be seen as having multiple valid forms. A weather forecast, a medical diagnosis, and a philosophical argument may all require different epistemic standards.

2. Expose human overconfidence in justification: Human justifications are often retrospective rationalizations, subject to bias and error.

3. LLMs raise the question: Can there be alien or non-sentient forms of knowledge? If so, epistemology must evolve to accommodate systems that know without understanding, or perform without belief. In sum, LLMs invite a revision of our epistemological intuitions: away from introspective justification and belief, and toward external performance, reliability, and generative capability.

DW: As a human, I take at least two things from this interchange. First, it’s easy to accept that an AI’s vast statistical model of the words used in the works it’s been trained on enable it to give a decent account of Western philosophical thought. But the final section on how AI’s knowledge-like ability should affect our understanding is not something widely discussed in its training materials. Its response thus seems like it must be doing original reasoning, even though the evidence so far is that it’s not.

Second, I suspect this type of dialogue with a non-conscious machine will become commonplace. This could profoundly change education, expertise, and knowledge itself. It doesn’t relieve us of thinking, but helps us to think more clearly and originally to confront challenges to our ideas … with the dangers and risks that KMWorld readers are well aware of.