Conferința Inferentialism and AI, Bogdan Dicher

Abstract: Large language models succeed at many linguistic and reasoning tasks, yet it remains unclear what, if anything, their internal workings amount to in semantic terms. This talk examines whether contemporary AI architectures can sustain a notion of meaning that is public, rule-bound, and inspectable—features typically expected by inferentialist theories of meaning.

I contrast two approaches. On the one hand, distributional semantics treats meaning as emerging from patterns of linguistic co-occurrence encoded in high-dimensional vector spaces. This view has been enormously successful in practice, but its semantic commitments are not entirely transparent: the structures that drive model performance are neither rule-governed nor readily interpretable as articulable commitments and entitlements.
On the other hand, inferentialism—especially in its proof-theoretic variants—ties content to the network of inferences a speaker is entitled to draw. This framework makes semantic norms explicit and publicly assessable, and it provides tools for distinguishing correct from defective inferential behaviour.

The core question is whether current AI systems instantiate anything like such normative, rule-structured inferential roles, or whether they merely approximate them statistically.

---

Conferința are loc în cadrul cursului de Filosofia minții de la specializarea Filosofie, dar este deschisă tuturor celor care doresc să participe (16 decembrie 2025, sala F. Ștefănescu Goangă, ora 18.00).