Scott Aaronson — Theoretical computer scientist, professor at UT Austin and director of its Quantum Information Center (formerly MIT), known for work in quantum computing and computational complexity.
The gist
In his second appearance on the podcast, Scott Aaronson explores big questions at the intersection of computation, physics, and mind. He discusses whether the universe is a simulation or computable, critiques Giulio Tononi's Integrated Information Theory and Roger Penrose's Godel-based views of consciousness, and assesses the Turing test, ELIZA, Eugene Goostman, and GPT-3. He then gives an accessible tour of complexity theory, including P, NP, PSPACE, BQP, and his beloved class SZK, and explains why most experts bet P does not equal NP. The conversation closes with reflections on the pandemic, institutional failure, cancel culture, and love.
Big reveals
- Aaronson introduces the 'pretty hard problem of consciousness' - determining which physical systems are conscious and to what degree.
- Tononi bit the bullet and agreed his theory implies an error-correcting code with high phi is more conscious than a person.
- Aaronson argues a giant uniform grid of XOR gates - essentially a blank wall - would be rated more conscious than a human by IIT.
- He explains Penrose needs the brain to be an uncomputable quantum-gravity computer, requiring five or six stacked speculations.
- He cites Scott Alexander's line that GPT-3 'ground up the entire internet into a slurry,' calling it a major advance over ELIZA.
- Asked where he bets, Aaronson says P is not equal to NP, putting only two or three percent odds on P equals NP.
- If P equals NP with an efficient algorithm, he says you could break all encryption, mine unlimited Bitcoin, and auto-solve the other Clay problems.
- He calls the failure of institutions like the CDC during the pandemic 'staggering' and one of the biggest failures in US history.
Things worth remembering
- Peter Shore joked that probing a black hole singularity might make the universe generate an overflow error and crash.
- IIT's measure phi has no real derivation - postulates are stated, then phi 'appears' as if pulled from a hat.
- The idea that Godel's theorem relates to consciousness was already old when Turing dismissed it in 1950.
- Aaronson names universality - that a few simple operations can express everything - as the most beautiful idea in computer science.
- Godel had conjectured an infinite hierarchy of programming languages, then read Turing's paper and admitted he was wrong.
- Internet encryption rests on the unproven belief that factoring is hard; best public algorithms are still exponential.
- Levin's universal search would solve NP problems efficiently if P equals NP, treating the brute-force prefix as a mere constant.
- A zero-knowledge proof can convince you two graphs are non-isomorphic via a wizard guessing correctly 100 times, revealing nothing about why.
- Lex urges patience and love around the upcoming election, warning anger could destroy the country and even the world.