Answers are grounded in this paper and cite their sources.
Readers of the philosophical literature on artificial intelligence (AI) will encounter an unexpected word, one that medicine has carried for two and a half millennia: pharmakon . Revitalised by Derrida and elaborated by Bernard Stiegler into a general theory of technology, the figure has lately been carried over to generative AI to capture the simultaneous enhancement and diminishment of cognition this technology produces in its users. The present paper argues that when philosophy borrows this word it should borrow it in full, in the sense it has acquired since Plato. The phármakon of contemporary clinical medicine is a relational object whose therapeutic or toxic character emerges in a specified encounter between substance, subject, and context. Stiegler’s pharmacology is itself relational, locating ambivalence in the process of transindividuation; but it does not provide an operational vocabulary for specifying how effects vary systematically between individual subjects. Clinical pharmacology provides such a vocabulary. Its accumulated learning — pharmacokinetic variability, pharmacogenetic polymorphism, idiosyncratic susceptibility, drug–drug interaction, therapeutic window, tolerance, withdrawal, and expectancy effects — has cognitive analogues in human interaction with large language models (LLMs) that have not yet been developed into a coherent pharmacological vocabulary and that early empirical research is beginning to measure. The paper proposes a reframing from the generic to the respondent pharmakon , not as a refutation of Stiegler’s pharmacology but as its specification at a level where clinical, pedagogical, and policy decisions can be made.
No takes yet. Share an insight, caveat, or question.
Haverkamp et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: