Ramsey's forgotten philosophy challenges AI's inductive foundations
A 1920s philosopher offers a new way to think about truth and uncertainty in AI systems.
Deep Dive
Frank Ramsey's ideas on truth, belief, and induction—explored in a new LessWrong essay—propose that 'truth' is redundant and that induction is best understood through betting behavior. This contrasts with Popper's falsificationism and the Cox‑Jaynes approach to Bayesian updating. The essay argues that adopting Ramsey's framework reshapes how we manage uncertainty, assess universal laws, and update probabilities without appealing to an independent truth correspondence.
Key Points
- Ramsey's redundancy theory of truth eliminates the need for a correspondence relation, simplifying how AI models represent truth values.
- Ramsey reinterprets induction as betting coherence, offering an alternative to Cox-Jaynes Bayesianism that is more action-oriented.
- The essay contrasts Popper's rejection of induction with Ramsey's pragmatic acceptance of non-deductive reasoning, relevant to AI's handling of universal laws.
Why It Matters
Ramsey's framework could reshape how AI systems reason about uncertainty, moving from metaphysical truth to actionable bets.