Scott Aaronson: Computational Complexity and Consciousness | Lex Fridman Podcast #130

Current physics fits well within the Church–Turing thesis—physical systems seem simulatable by Turing machines—yet a perfect simulation is by definition empirically indistinguishable from reality, making the simulation hypothesis scientifically inert unless we can find and exploit ‘bugs’ in the laws of nature.

Scott’s “pretty hard problem” is to specify, from physical or informational properties alone, which systems (brains, AIs, animals, fetuses) are conscious and to what degree; any candidate theory that says simple error-correcting circuits or blank-wall–like grids are ‘more conscious’ than humans is, in his view, talking about the wrong thing.

IIT’s Phi measure can be made enormous using simple, structurally dense circuits that have no plausible claim to consciousness, and its derivation is more hand-wavy than axiomatic—so Scott argues that while it’s bold and ambitious, it doesn’t match our pre-theoretic target concept of consciousness.

For Penrose to be right, we’d need new quantum-gravitational, fundamentally uncomputable physics; that physics would have to matter in warm, wet brains; quantum mechanics would need objective collapse; and conscious intentions would have to bias collapse outcomes—all driven by an argument from Gödel’s incompleteness that most experts find unconvincing.

Scott would bet heavily that problems whose solutions are easy to verify (NP) are not all easy to solve (P); but if P = NP with a practical algorithm, modern cryptography would collapse, optimal neural networks and proofs to major open math problems could be found algorithmically, and computational practice and theory would be transformed.