AI Safety

LessWrong post maps all decision theories, reveals entangled axes

A 50-minute read that unifies Newcomb, FDT, and UDT into a single provably entangled grid

Deep Dive

In a LessWrong post, Ihor Kendiukhоv proposes a map of decision theories built on three separate axes: what we suppose when considering a candidate action, how

Key Points
  • The post decomposes decision theory into three semi-orthogonal axes: suppositions about actions, gamble scoring, and pre- vs post-information choice.
  • It maps existing theories like EDT, CDT, FDT, and Wei Dai's UDT onto a grid, with theorems proving the axes are entangled.
  • Includes an AI-generated technical appendix with formal proofs and open problems, explicitly marked as 'no guarantee.'

Why It Matters

A unified map of coherent decision theories could sharpen AI alignment's foundations for agent design and safety.

📬 Get the top 10 AI stories daily