What Capable Agents Must Know: Why AI Consciousness May Be an Inevitable Byproduct of Capability

Short summary

A UAI 2026 paper proves selection theorems showing that capable agents must necessarily develop world models, belief-like memory, emotion primitives, and representational convergence. The author argues these architectural prerequisites may constitute a bridge toward first-person subjective experience, suggesting consciousness could be an inevitable byproduct of AI capability rather than requiring special architectural conditions. The second arrow—from these cognitive primitives to actual consciousness—remains unproven but is supported by preliminary evidence from neuroscience and modern LLM behavior.

• •Selection theorems prove capable agents necessarily develop world models, belief-like memory, emotion primitives, and representational convergence
• •Author argues these prerequisites may bridge toward consciousness, updating from a previous view that intelligence and consciousness are orthogonal
• •The link from cognitive primitives to subjective experience remains unproven but is supported by preliminary neuroscience and LLM evidence

Generated with AI, which can make mistakes.