Lex Fridman PodcastScott Aaronson: Quantum Computing | Lex Fridman Podcast #72
CHAPTERS
- 5:03 – 8:40
Why philosophy matters for scientists (and why it often gets sidelined)
Lex asks why technical experts should care about philosophy, and Scott reframes philosophy as the domain of the biggest questions—often too big for direct progress. Scott argues that science makes headway by narrowing questions into tractable forms, while still reshaping our understanding of the original philosophical problems.
- •Philosophy as the study of the biggest, hardest-to-answer questions
- •Scientific progress comes from focusing on narrower, answerable sub-questions
- •Technical work can indirectly transform philosophical debates
- •Why many scientists prioritize tractable problems over metaphysical ones
- 8:40 – 14:07
Turing vs. Wittgenstein: formal systems, meaning, and real-world stakes
Scott describes the fascinating 1939 interaction between Alan Turing and Ludwig Wittgenstein, illustrating the tension between dismissing formalism and recognizing its practical power. The episode highlights how formal systems became essential once computation and engineering entered the picture.
- •Turing’s philosophical engagement (Mind paper, Turing test)
- •Wittgenstein’s skepticism about formal systems’ relevance
- •Turing’s pragmatic argument: inconsistency can break real systems (bridges)
- •How computing made formalism materially consequential
- 14:07 – 16:20
Turning unanswerable riddles into solvable ones: the “Q prime” framework
Lex brings up Scott’s idea of replacing an unanswerable philosophical question (Q) with a related, tractable scientific/mathematical version (Q′). Scott gives historical examples where reframing—rather than direct answering—produced real progress.
- •Q → Q′ as a strategy for intellectual progress
- •Turing’s move: ‘Can machines think?’ → Turing test framing
- •Gödel’s move: limits of reasoning → formal incompleteness statements
- •Why reframing often beats debating definitions
- 16:20 – 24:46
Free will as predictability: brain models, quantum effects, and measurement limits
Scott explores a Q′ for free will: how well can a person’s behavior be predicted in principle, consistent with physics? The discussion touches on chaos, thermal noise, quantum randomness, and the measurement disturbance problem when trying to model a brain precisely.
- •Free will reframed as limits on predicting behavior
- •Chaotic amplification of microscopic/quantum events in neural processes
- •Randomness vs meaningful freedom: ‘probabilistic prediction’ still feels threatening
- •Quantum measurement as inherently disturbing—scanning may alter the system
- 24:46 – 29:45
Consciousness, AI, and how future technologies could change the debate
They connect free will and consciousness, noting the ‘hard problem’ has resisted progress since antiquity. Scott argues that even without solving it, technologies like human-level AI or a perfect behavior-predictor could fundamentally change how we discuss these questions.
- •Hard problem of consciousness as a recurring philosophical bottleneck
- •AI as a potential ‘Q′’ that reshapes consciousness discussions
- •Prediction machines would alter lived experience of agency
- •Demonstration vs essence: behavior-based tests and their limits
- 29:45 – 37:44
Quantum computing basics: amplitudes, superposition, and interference (not ‘parallel universes’)
Scott gives a foundational explanation of quantum computing as computation built from quantum mechanics—especially interference. He emphasizes why the common ‘tries all answers in parallel’ story is misleading: measurement yields randomness unless interference is engineered to amplify correct outcomes.
- •Quantum mechanics as a generalization of probability via complex amplitudes
- •Measurement converts amplitudes to probabilities (squared magnitudes)
- •Double-slit interference as the core intuition
- •Quantum algorithms = choreographing interference so wrong answers cancel
- 37:44 – 41:28
Qubits as information: physical implementations vs abstract computation layers
The conversation shifts to information: bits, qubits, and how different physical systems can realize qubits while sharing the same computational ‘logic.’ Scott notes that today hardware imperfections leak into the abstraction, but error correction aims to restore clean layers.
- •Bit vs qubit as fundamental units of information
- •Multiple qubit implementations (superconducting circuits, nuclear spins, etc.)
- •Programming abstraction exists in principle, but noise currently breaks it
- •Goal: error-corrected quantum computers that decouple software from hardware
- 41:28 – 47:46
Decoherence and quantum error correction: why scaling is so hard
Scott explains decoherence as unwanted entanglement with the environment—effectively a continuous ‘measurement’ that destroys quantum information. He outlines the key 1990s breakthrough: fault tolerance and error correction make scalable QC possible in theory, but require extremely low error rates and massive overhead.
- •Decoherence as information leakage to the environment
- •Isolation vs control tension: protect qubits but still manipulate them
- •Quantum error correction/fault tolerance enables reliability from noisy parts
- •Overhead realities: logical qubits require many physical qubits (potentially millions)
- 47:46 – 51:03
Engineering path forward: NISQ era, investment, and theoretical breakthroughs
They discuss the current ‘noisy intermediate-scale quantum’ (NISQ) phase—likened to early vacuum-tube computing—where devices can outperform classical simulation on narrow tasks but lack full scalability. Scott suggests progress will likely require a mix of engineering improvements, theoretical advances, and sustained investment.
- •NISQ as early, noisy stage before true fault tolerance
- •Why a ‘Manhattan Project’ scale spend could accelerate progress (in principle)
- •Likely need both hardware advances and new theory to cut costs
- •Near-term goal: useful advantage beyond classical capabilities
- 51:03 – 56:41
Moore’s Law, universality, and fundamental physical limits
Lex asks philosophically why humans keep producing CPU breakthroughs; Scott calls our era historically special due to universal programmable machines and economic pressure. He argues Moore’s Law can’t continue indefinitely, citing ultimate physics constraints (even black-hole limits at absurd extremes).
- •Computing progress concentrates across civilization due to economic incentives
- •Classical universality shifts progress to ‘numbers’ (speed, memory, parallelism)
- •Moore’s Law must end due to physical constraints
- •Quantum computing as a genuine shift beyond ‘just more transistors’
- 56:41 – 1:02:27
Quantum supremacy defined: a well-defined task where quantum wins (usefulness optional)
Scott defines quantum supremacy as the first clear demonstration that a quantum device can do a specified task faster than any known classical method. He clarifies that quantum computers don’t compute the uncomputable; they change what’s efficiently computable (polynomial vs exponential scaling).
- •Supremacy as a milestone: quantum advantage on a precise benchmark
- •Not necessarily useful—must be well-defined and verifiable
- •Quantum doesn’t break Turing computability (no halting problem miracles)
- •Efficiency as polynomial scaling; why scaling behavior is central
- 1:02:27 – 1:12:15
How Google’s supremacy experiment worked: sampling, verification, and classical simulation limits
Scott explains why sampling problems are ideal for near-term demonstrations and walks through Google’s random-circuit sampling with 53 qubits. He describes verification via the linear cross-entropy benchmark and the subtlety: you can’t prove impossibility of classical spoofing, but you can build strong complexity-theoretic evidence and test known simulators.
- •Sampling tasks: many valid outputs, but with a target distribution
- •Random circuits produce structured ‘garbage’ shaped by interference
- •Verification via cross-entropy requires heavy classical computation
- •Hardness arguments rely on reductions and complexity assumptions; plus empirical checks
- 1:12:15 – 1:17:14
Quantum threats to cryptography: Shor’s algorithm, timelines, and post-quantum migration
Lex brings up claims that quantum makes all code crackable; Scott says that’s premature. He distinguishes today’s devices from the millions-of-qubits, error-corrected machines needed to run Shor’s algorithm at cryptographically relevant scales and points to post-quantum cryptography efforts (notably lattice-based systems and NIST standardization).
- •Shor’s algorithm breaks RSA/discrete log on scalable fault-tolerant QCs
- •Current ‘supremacy’ devices are far from cryptographically relevant capability
- •‘Harvest now, decrypt later’ risk motivates early migration
- •Post-quantum cryptography candidates and NIST standardization process
- 1:17:14 – 1:33:41
Practical quantum value: chemistry/materials simulation, cautious optimism about QML, and life meaning
Scott argues the most promising long-term application is simulating quantum systems to accelerate chemistry and materials discovery, possibly impacting industries with a few key breakthroughs. He then critiques hype around quantum machine learning, highlighting ‘dequantization’ results (Ewin Tang) and insisting on demonstrated classical-vs-quantum speedups; the episode closes with Scott’s reflections on meaning, relationships, and making the world better.
- •Most plausible major impact: quantum simulation for chemistry/materials/drugs
- •Near-term uncertainty: what’s feasible with 100–200 noisy qubits vs needing logical qubits
- •QML hype vs reality: focus on provable/credible speedups; Grover is only quadratic
- •Dequantization case study (Ewin Tang) and the call for rigor; closing thoughts on meaning of life
