Interview
Scott Aaronson: Quantum Computing | Lex Fridman Podcast #72
- Plans to revisit computational complexity theory and the Complexity Zoo wiki in future discussions, with current focus on quantum computing.
- Anticipates that the existence of human-level or overtaking AIs would fundamentally alter the discourse on the hard problem of consciousness, while the potential realization of a "prediction machine" would transform concepts of free will, even if metaphysical questions remain unanswered.
- Projects that error-corrected quantum computers envisioned in the 1990s could initially cost a trillion dollars or more, though further theoretical breakthroughs are expected to reduce these costs.
- Predicts reaching fundamental physical limits on processor miniaturization imposed by quantum gravity, specifically noting that a computer operating at 10^43 operations per second would generate enough energy to collapse into a black hole.
- Foresees a race within the next decade to achieve useful quantum simulations using 100-to-200 qubit systems, citing potential applications in material science, chemistry, and nuclear physics.
- References a Microsoft study suggesting that roughly 100 qubits could already yield new insights into the chemical reactions involved in fertilizer production.
- Expects a future migration of the internet to post-quantum cryptography standards currently being developed via NIST competitions to upgrade browsers and routers against quantum threats.
- Predicts that soon, classical computers will lose the ability to verify quantum sampling results, as quantum devices will surpass the capabilities of the largest classical machines using significantly less time.
- States that if a fast classical algorithm can spoof these experiments, it must differ radically from known algorithms.
- Outlines a simultaneous multi-directional push to improve qubit fidelity (which has improved by orders of magnitude over the last decade or two), design lower-overhead error-correcting codes, and develop more efficient quantum algorithms for simulation.
- Notes that while many proposed quantum machine learning algorithms have been "de-quantized," the field currently has evidence of modest speedups, with the potential for future exponential speedups remaining a major open question.
- Emphasizes that error mitigation strategies are being utilized to achieve useful speedups in the near term while the ultimate goal remains demonstrating definitive speedup over classical methods.
- Concludes with a commitment to engage with global crises regarding climate change and resurgent authoritarianism.