Scott Aaronson on Quantum Computing, OpenAI, and Human Safety

Scott Aaronson is the David J. Bruton Jr. Centennial Professor of Computer Science at the University of Texas at Austin and director of its Quantum Information Center. Previously, he was on the faculty at the Massachusetts Institute of Technology. He studied at Cornell and University of California, Berkeley, and did postdocs at the Institute for Advanced Study as well as the University of Waterloo. His first book, Quantum Computing Since Democritus, was published in 2013 by Cambridge University Press. Aaronson has written about quantum computing for Scientific American and the New York Times, and writes a popular blog. He’s received the National Science Foundation’s Alan T. Waterman Award, the United States PECASE Award, and MIT’s Junior Bose Award for Excellence in Teaching.

Aaronson’s research focuses on the capabilities and limits of quantum computers and more generally on computational complexity and its relationship to physics. Within the context of the It from Qubit Collaboration, Aaronson is extremely interested in the interplay between computational complexity and quantum gravity. This has involved studying the computer-science aspects of the Harlow-Hayden argument, which attempts to apply complexity theory to the notorious “firewall paradox” in black hole information, as well as working with Leonard Susskind to understand the growth of quantum circuit complexity in systems arising from the AdS/CFT correspondence.

Relevant links:

Timestamp:

00:28 -- Post-interview remarks and introduction by Razo
06:08 -- Interview begins
07:19 -- How Aaronson feels about his teaching mission
07:37 -- Aaronson describes his blog and its original purpose
08:58 -- Aaronson's "reactive" approach to correcting errors
13:08 -- The reality of quantum versus classical computing performance
13:50 -- The classes of problems where quantum computers are better
17:41 -- Combinatorial optimization as the holy grail of quantum speedup
18:53 -- The relatively modest "quadratic" speed up of Grover's algorithm
21:31 -- Can quantum computers solve the P versus NP problem?
22:37 -- The story that Aaronson has been clarifying for 20 years
23:58 -- Is Aaronson optimistic or pessimistic about quantum computing?
26:48 -- Measurement of a quantum system as a destructive and limiting process
28:27 -- The hardest part of quantum algorithm research - beating classical
29:52 -- The problem of quantum algorithms being "dequantized"
30:15 -- Aaronson's student, 18-year-old Ewin Tang, makes important breakthrough
31:57 -- The hope of some for quantum help in optimization / simulated annealing
36:19 -- Razo on recent efforts to connect quantum theory to social science
38:48 -- Aaronson on past attempts to connect quantum theory and social science
41:00 -- Aaronson objects to using analogies that don't provide causal explanations
42:45 -- The universality of mathematical ideas and applied linear algebra
46:44 -- Razo responds to the Aaronson's objection about analogies
49:26 -- Aaronson rejects math analogies that complicate things needlessly
50:55 -- Razo clarifies that the proposed analogies actually solve real problems
51:25 -- Aaronson explains that quantum ideas have led to past breakthroughs
52:35 -- What is Aaronson working on now? (How OpenAI approached him)
54:04 -- IP = PSPACE as an interesting analogy for AI safety
55:22 -- How do we get a human verifier to check an AI prover?
56:15 -- Aaronson talks about not being able to move past PSPACE as an analogy
57:09 -- Aaronson's work on statistical watermarking to fight misuse of ChatGPT
59:00 -- How to protect humans from an AI that is much smarter than humans?
59:59 -- Razo on how a human in the loop leads to voting theory

[eismscience]

Timestamp:

00:28 -- Post-interview remarks and introduction by Razo
06:08 -- Interview begins
07:19 -- How Aaronson feels about his teaching mission
07:37 -- Aaronson describes his blog and its original purpose
08:58 -- Aaronson's "reactive" approach to correcting errors
13:08 -- The reality of quantum versus classical computing performance
13:50 -- The classes of problems where quantum computers are better
17:41 -- Combinatorial optimization as the holy grail of quantum speedup
18:53 -- The relatively modest "quadratic" speed up of Grover's algorithm
21:31 -- Can quantum computers solve the P versus NP problem?
22:37 -- The story that Aaronson has been clarifying for 20 years
23:58 -- Is Aaronson optimistic or pessimistic about quantum computing?
26:48 -- Measurement of a quantum system as a destructive and limiting process
28:27 -- The hardest part of quantum algorithm research - beating classical
29:52 -- The problem of quantum algorithms being "dequantized"
30:15 -- Aaronson's student, 18-year-old Ewin Tang, makes important breakthrough
31:57 -- The hope of some for quantum help in optimization / simulated annealing
36:19 -- Razo on recent efforts to connect quantum theory to social science
38:48 -- Aaronson on past attempts to connect quantum theory and social science
41:00 -- Aaronson objects to using analogies that don't provide causal explanations
42:45 -- The universality of mathematical ideas and applied linear algebra
46:44 -- Razo responds to the Aaronson's objection about analogies
49:26 -- Aaronson rejects math analogies that complicate things needlessly
50:55 -- Razo clarifies that the proposed analogies actually solve real problems
51:25 -- Aaronson explains that quantum ideas have led to past breakthroughs
52:35 -- What is Aaronson working on now? (How OpenAI approached him)
54:04 -- IP = PSPACE as an interesting analogy for AI safety
55:22 -- How do we get a human verifier to check an AI prover?
56:15 -- Aaronson talks about not being able to move past PSPACE as an analogy
57:09 -- Aaronson's work on statistical watermarking to fight misuse of ChatGPT
59:00 -- How to protect humans from an AI that is much smarter than humans?
59:59 -- Razo on how a human in the loop leads to voting theory

[aminam]

Better to level listen to Scott Aaronson carefully, this man has valuable thoughts.

[suzannecarter]

I'm not a scientist, but I do try to keep up with developments. I know many people seem to believe that quantum computers will be required to solve the really hard "complexity theory" problems. Is it possible that rapid developments in AI could change that, i.e., find methods for solving complexity science problems using classical computers?

[peteunderdown]

Scott is so polite. But that was painful.

[aminam]

This man isn’t Hallucinator (realistic).