Causation as Sign Action: Semiotic Empiricism and the Causal Complexity of Large Language Models – M. Beatrice Fazi (University of Sussex)

Abstract:

Can large language models (LLMs) engage with causation, or are they confined to detecting correlations in data? This talk argues that addressing this question requires rethinking what causation itself is and how it becomes intelligible. Drawing on Charles Sanders Peirce’s triadic theory of signs, the talk develops the position that causation is semiotically constituted: causal relations become meaningful not through experiential grounding alone but through irreducibly triadic sign processes involving signs, objects, and interpretants. Albert Michotte’s “launching effect” experiment provides a key departure point, demonstrating that causal perception is already semiotic rather than a matter of direct apprehension or Humean inference. This reframing challenges Judea Pearl’s influential “ladder of causation”—which positions LLMs as processors of correlations incapable of causal reasoning—while also taking seriously Pearl’s own acknowledgment that causal information is embedded in the training texts on which these models operate. The talk proposes “semiotic empiricism” as a framework in which experience is structured by meaning-making rather than preceding it, and examines its consequences for understanding how LLMs might participate in causal reasoning as semiotic machines, thus moving the debate beyond both dismissive critique and anthropomorphic overclaiming.

This event is part of our summer term 2026 Lecture Series Digital Complexity: Beyond Human Understanding.

If you would like to attend, please register with events@khk.rwth-aachen.de.