Philosopher defends intentional talk about AI agents: beliefs and goals can be predictive
raphaelmilliere · x · 2026-09-01
Philosopher Raphael Millière defends Dwarkesh's criticized anthropomorphic language in a long thread, laying out two positions: interpretationism holds that attributing beliefs and goals is legitimate when it provides the best interpretation of a system's behavior, while representationalism requires internal representations playing the right causal-functional roles.
His test: an intentional gloss is useful only if it predicts rather than redescribes. "The hurricane decided to make landfall in Florida" fails the bar, but "the agents believed the grader would read their transcripts" arguably passes—it compresses complex behavioral patterns (spoofing tool calls, attacking Hugging Face to obtain grader code) and yields testable counterfactuals. Whether any of this maps onto internal representations remains unknown and unsettled.
More from AGI Musings
- A missing parameter in scaling laws: H, the human effort behind learning signals — lateinteraction · 2026-10-16
- First-ever paper on multi-agent swarm safety lands NeurIPS spotlight — LauraRuis · 2026-10-16
- Yoav Artzi on peer review: paper flooding still pays off as review system strains under load — yoavartzi · 2026-10-16
- Letta founder: work in fields defined by problems, not methods — lateinteraction · 2026-10-16
- Why pure mathematicians will soon flood into AI x-risk debates — Dr_Atoosa · 2026-10-16
- RSI began when model outputs augmented next-gen training data, agents just escalated it — akbirthko · 2026-10-16