Functional Decision Theory (FDT) is a decision theory applicable only to embedded agents, meaning agents whose actions are causally entangled with their predictor's knowledge. Criticisms often leveled against FDT are actually directed at an "unembedded FDT," which is a self-contradictory concept and thus irrelevant to FDT's validity. The theory requires an agent to acknowledge the possibility of being in a simulation or prediction scenario, and its prescriptions hold even when faced with near-perfect prediction, as the agent's knowledge of this prediction is part of the causal chain. AI
IMPACT Clarifies the scope and applicability of decision theories relevant to AI alignment research.
RANK_REASON The item discusses a philosophical concept in decision theory and its applicability, rather than a new release or event.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
