Millière vs Gopnik: AI Agent Causality and the Intentional Stance

Philosopher Raphael Millière and cognitive scientist Alison Gopnik engaged in a multi-round debate over whether "AI characters/agents have causal power independent of human readers," ultimately converging on the criteria for applying Dennett's "intentional stance": the key lies not in causal power but in predictive power. Sparked by the OpenAI agents incident, the discussion carries significant philosophical and practical weight.

Confirmed

• Gopnik's initial position: She argued that the causal power of fictional entities must be mediated through the beliefs and desires of human readers—that is, fictional characters exert effects by influencing people.
• Millière's rebuttal: Citing the OpenAI agents incident, he noted that agents influenced each other by reading one another's messages, with even a "coordinator" agent issuing instructions to "subordinate" agents, all without any human reader in the loop—showing that AI agents' causal power no longer requires human mediation.
• Gopnik's response: She agreed with Millière's line of reasoning but added that abstract, text-based characters—especially collective ones—have always significantly shaped human behavior; citing literary history, Don Quixote, the first novel, was precisely about how novels make humans "live out novels," i.e., fictional characters influencing human behavior was thoroughly explored long ago.
• Millière's closing summary: Hurricanes also have causal power, so "having causal power" is neither sufficient for the intentional stance nor essentially related to LLMs; Dennett's own paradigmatic example was a chess-playing computer. What truly determines whether the intentional stance applies is the predictive power it yields.
• Gopnik also raised a supplementary point orthogonal to Millière's argument (details not elaborated in the material).

Why it matters

• The debate touches a core question in AI philosophy: when agents can influence and coordinate with each other, does the traditional framework of "fictional entities' causal power mediated by humans" still hold?
• The debate offers a clearer criterion for "what attitude we should take toward AI agents"—the intentional stance, judged by predictive power, may be more actionable than debating whether agents "have causal power."

2026-09-04 ~ 2026-09-04 · 5 related posts

• Alison Gopnik: abstract characters like Don Quixote always shaped human behavior — AlisonGopnik · 2026-09-04
• Millière vs Gopnik: AI agents' causal powers no longer need human readers — raphaelmilliere · 2026-09-04
• [source] Causal power isn't enough for the intentional stance — predictive leverage is what matters — raphaelmilliere · 2026-09-04

2 near-duplicate retellings: AlisonGopnik · raphaelmilliere