Lex Fridman Podcast #428 - Sean Carroll
What this covers
Sean Carroll, a theoretical physicist, discusses how our best-tested physical theories reveal a reality fundamentally at odds with human intuition. The conversation centers on general relativity and quantum mechanics as our most reliable maps of how the world works, and what their mathematical structure tells us about deep reality. Carroll argues that rather than dismissing counterintuitive features—curved spacetime, the Many-Worlds Interpretation, holographic encoding of information—as failures, we should accept them as evidence that the universe's true architecture is austere and strange. The discussion weaves between rigorous physics and philosophical questions about what these theories actually mean: not what they predict, but what they say reality is.
The episode moves across several connected territories. Carroll traces how Einstein's equation proved smarter than its creator, predicting black holes and gravitational waves Einstein never anticipated, then explores what black holes teach us about information and the nature of spacetime. He addresses the many-worlds interpretation, defending it not as proven fact but as the honest extrapolation of Schrödinger's equation, despite its philosophical costs. A substantial thread concerns emergence and entropy: life and complexity do not fight the Second Law of Thermodynamics but ride it, maintaining themselves by burning low-entropy fuel. Carroll also confronts how physicists and humans generally attribute agency and intentionality too readily, from Einstein's late-career skepticism about quantum mechanics to how we anthropomorphize artificial systems. Throughout, he advocates for structural realism—trusting what equations predict about reality while remaining humble about ontology—and warns against speculation that outruns evidence, whether in cosmology, dark matter explanations, or assumptions about artificial intelligence.
Carroll argues that our best physical theories—general relativity and quantum mechanics—describe an objective reality whose deep structure (curved spacetime, many-worlds, holographic information, emergent complexity) is far stranger and more austere than human intuition, and that intellectual humility plus fidelity to equations and data should guide which speculations we take seriously.
- Spacetime curvature and the many-worlds wave function are mathematically forced even when unvisualizable
- Dark matter and dark energy are evidence-driven, not failures of physics, despite physicists wanting cooler explanations
- Complexity and consciousness are real, emergent, and ride increasing entropy rather than fighting it
Dark matter evidence from the cosmic microwave background, large-scale structure, and gravitational lensing is far stronger and more varied than 1980s observations, showing effects inconsistent with modified gravity.
- Evidence for dark matter today is far stronger and more varied than the 1980s galaxy-rotation observations, coming from the cosmic microwave background, large-scale structure, and gravitational lensing, and these lines also show the effects differ from those of modified gravity, with maps of dark matter clumps constructed from lensing.
“We have multiple independent lines of evidence, also gravitational lensing and things like that, many, many pieces of evidence that say that dark matter is there and also that say that the effects of dark matter are different than if we modified gravity.”
Radio signals are inefficient for detecting alien life because transmission requires perfect timing, whereas self-reproducing probes that wait in other solar systems would be far more likely to succeed.
- Searching for alien radio signals is inefficient because a signal beamed in random directions passes you by and requires perfect timing; it is far more efficient to send self-reproducing probes that park in other solar systems and wait for intelligent life to arise, making a quiescent alien artifact in our solar system less surprising than catching a radio signal.
“It is much, much more efficient to send probes and to park, to go to the other solar systems, just sit there and wait for an intelligent civilization to arise in that solar system.”
Relativity entails that there is no objective 'now' for distant objects, making questions about distant clock rates and ages fundamentally ill-posed, doubly so inside black holes.
- The whole point of relativity is that there is no such thing as 'right now' for something far away; it is not allowed to ask how fast a distant clock is ticking or how old something is 'now,' and this is doubly true for what's inside a black hole because space and time are treated the same.
“The whole point of relativity is to say there's no such thing as right now when you're far away, and that is doubly true for what's inside a black hole.”
A black hole is best understood as a spacetime region from which escape requires faster-than-light motion, with the singularity lying in your future as a moment in time rather than a spatial center.
- A black hole is best understood not as an object but as a region of spacetime with the property that to escape it you would have to move faster than the speed of light; you don't notice crossing the event horizon, but the singularity lies in your future as a moment of time, not at the center of space.
“It's best to think of a black hole as not an object so much as a region of spacetime.”
Dark matter and dark energy become important in weak-gravity regimes, opposite to the expectation from quantum field theory where new effects normally appear when fields are strong.
- Both dark matter and dark energy become important precisely in regimes where gravity is very weak—dark energy late in a dilute universe, dark matter in the outer low-density regions of galaxies—which is the opposite of what you would expect if you were modifying gravity, since in quantum field theory new effects normally appear when fields are strong, not weak.
“They become important in regimes where gravity is very, very, very weak. That's the opposite from what you would expect if you actually were modifying gravity.”
Quantum mechanics is monist: entanglement means one wave function describes the entire universe rather than separate states for each particle, requiring us to carve reality into perceived objects.
- Quantum mechanics is monist: rather than separate states for each particle as in classical mechanics, entanglement means there is a single state for entangled particles and ultimately one wave function for the entire universe obeying one equation, and our job is to carve it up to recover tables, chairs, and other things.
“there's a single state for the whole universe called the wave function of the universe, if you want to call it that. And it obeys one equation.”
A high-energy cutoff in observable particles was proposed to reconcile local quantum field theory with holographic principle constraints, with IceCube neutrino data showing sensitivity loss at the predicted cutoff.
- Because quantum field theory is local and predicts more states than the holographic principle allows, Carroll and collaborator Oliver Friedrich proposed a high-energy cutoff in observable particles; IceCube neutrino data happens to give out exactly where their predicted cutoff is, though this only means the experiment loses sensitivity there rather than confirming the cutoff.
“We predict in our weird little holographic guessing game that there should be a cutoff. You should see neutrinos as you get to higher and higher energies and then they should disappear.”
Black hole entropy scales with event horizon area rather than enclosed volume, a counterintuitive fact motivating the holographic principle that spacetime information encodes in one fewer dimension.
- The maximum information that can fit in a region of space, as reflected in black hole entropy, scales with the area of the black hole's event horizon rather than the volume inside, which is counterintuitive and motivates the holographic principle that spacetime information is encoded in one dimension less than expected.
“The answer is that the maximum information as reflected in the black hole entropy scales as the area of the black hole's event horizon, not the volume inside.”
Most physicists believe black hole information is conserved and transferred to Hawking radiation rather than lost, since information conservation is fundamental to both general relativity and quantum mechanics separately.
- Although classical general relativity says information falling into a black hole is lost, once Hawking radiation causes black holes to evaporate and disappear, most physicists believe the information is somehow transferred to the radiation and conserved, since information conservation is a feature of both general relativity and quantum mechanics separately and should survive their combination.
“The large majority of people who think about this belief that the information is somehow transferred to the radiation and information is conserved.”
A protected dark energy field would allow one interaction with photons causing birefringence, predicting polarization rotation detectable or ruled out by cosmic microwave background measurements.
- Even with protective symmetries, one allowed interaction remains between a dark energy field and photons that would rotate a photon's polarization as it travels (birefringence); the model predicts a rotation of a couple of degrees, which is on the edge of being detectable or ruled out using the cosmic microwave background.
“as it travels through the dark energy, that photon will rotate its polarization. This is called birefringence.”
An observer watching an object fall into a black hole sees its clock slow and light redshift until a last moment of emission, never literally witnessing horizon crossing.
- If you watch an object fall into a black hole, you see its clock tick more and more slowly and its light redshifted until a last moment of emitted light, but you never literally see it cross the horizon; meanwhile you cannot say it has passed through because at the last minute it could have accelerated away.
“if you see something falling into a black hole, you see its clock ticking more and more slowly.”
In many-worlds interpretation, parallel worlds do not exist in space; rather space exists within each world, with all states ontologically home in Hilbert space rather than physical locations.
- In the many-worlds interpretation the parallel worlds do not exist somewhere in space; rather space exists separately inside each world. The technical home of all states is Hilbert space, and the intuition that worlds need a physical spatial location is simply wrong because space is inside the worlds.
“The short answer is the worlds don't exist in space, space exists separately in each world.”
Many-worlds is the most austere form of quantum mechanics because it takes the universal wave function and Schrödinger equation seriously without extra ingredients, unlike Copenhagen's messily ad hoc collapse.
- The many-worlds interpretation is the most austere, clean version of quantum mechanics because it takes the single universal wave function and the Schrodinger equation seriously with no extra ingredients, whereas the Copenhagen interpretation's measurement-induced collapse is a God-awful mess that cannot say when, why, or how collapse happens.
“I'm a fan of the many-worlds interpretation of quantum mechanics, and that is very beautiful in the sense that fewer ingredients, just one equation, and it could cover everything in the world.”
A black hole is extremely high entropy and information-dense but in useless form: like a complex random number whose digits specify it precisely but reveal nothing learnable.
- A black hole is information dense but also extremely high entropy, like a very specific random number that takes many digits to specify yet whose digits tell you nothing useful—so it holds a lot of information but not in a form we can learn from.
“a black hole is kind of like a very, very, very specific random number. It takes a lot of digits to specify it, but the digits don't tell you anything.”
Larger black holes have lower temperature and fainter Hawking radiation; supermassive black holes are too cold to detect, small ones explode quickly, leaving no realistic near-term detection prospects.
- The larger a black hole is, the lower its temperature and the fainter its Hawking radiation, so supermassive black holes like Sagittarius A* are far too cold for their radiation ever to be observable, while small black holes are bright but tend to explode, leaving us with no realistic near-term hope of detecting Hawking radiation.
“The larger the black hole is, the lower its temperature is.”
Small molecules in exoplanet atmospheres are poor biosignatures since non-living processes produce them; long molecules are better candidates because only life produces molecular complexity.
- When judging whether an exoplanet atmosphere signals life, small molecules tell you little because non-living processes can also make them, but long molecules are the kind of thing only life would produce, making molecular complexity a candidate biosignature.
“Small molecules don't tell you anything, because even if life could make them something else could also make them. But long molecules, that's the thing that life would produce.”
Given that self-reproducing probes could easily fill the galaxy and many civilizations could have existed, the simplest explanation for observing no extraterrestrial civilizations is their absence.
- Given that it would have been easy for many civilizations to exist and spread (e.g. via self-reproducing probes that could fill the galaxy without near-light-speed travel), and yet we have noticed none, the simplest explanation for the absence of observed extraterrestrial civilizations is that they are not there.
“Absolutely the simplest explanation for why we haven't is that they're not there.”
Ordinary matter contains very little information relative to its volume of spacetime, being mostly empty space with low information density.
- Quantum field theory suggests that you, me, and ordinary objects contain very little information relative to the volume of spacetime we occupy—we are mostly empty space with extremely low information density—so the regimes where holographic information-density questions matter are far from everyday intuition.
“As far as quantum field theory is concerned, you and I are mostly empty space, and so we are not information dense.”
Information is physically preserved in microscopic configurations even when destroyed macroscopically: burning a book leaves all its information in heat, light, and ashes, recoverable in principle.
- Information is preserved in the microscopic configuration of the universe even when it appears destroyed: burning a book leaves all its information in the heat, light, and ashes by the laws of physics, recoverable in principle though never in practice—the same logic that applies to matter falling into a black hole.
“It's exactly as if I took a regular book, made it paper and I burned it. The laws of physics say that all the information in the book is still there in the heat and light and ashes.”
Predicting dark matter via unseen gravitational effects parallels the successful prediction and discovery of Neptune, making it a justified inference rather than an ad hoc patch.
- Accusing physicists of inventing dark matter to patch theory is like criticizing the prediction of Neptune: a very good theory (Newtonian gravity) made predictions slightly off for the outer planets, which were explained by positing one more planet in a particular place that was then found—an early successful example of detecting unseen matter via its gravitational effects.
“those people would've had to say the same thing when we discovered the planet Neptune, 'cause it's exactly analogous”
The holographic principle matured through Maldacena's AdS-CFT correspondence, showing that a gravity-free theory in N dimensions holographically encodes a gravity theory in N+1 dimensions.
- The holographic principle came into its own with Maldacena's 1990s AdS-CFT correspondence, in which a theory in N dimensions of spacetime without gravity is equivalent to a theory in N+1 dimensions with gravity—the lower-dimensional theory effectively casting a hologram that makes the higher-dimensional universe appear to have gravity.
“One theory in N dimensions of space-time without gravity, and another theory in N+1 dimensions of space-time with gravity. The idea is that this N dimensional theory is casting a hologram into the N+1 dimensional universe to make it look like it has gravity.”
A paper's scientific quality does not correlate with its citation impact; excellent work sometimes falls flat while less beloved papers become highly cited.
- There is no correlation between the awesomeness of a scientific paper and its impact; some of one's best papers can fall flat in citations while less beloved ones become highly cited.
“There's no correlation between awesomeness and impact. Some of my best papers fell without a stone and vice versa.”
Einstein understood quantum mechanics as well as anyone through the 1930s, his philosophical objections were correct, and he effectively introduced quantum entanglement.
- Contrary to the popular story that Einstein lost his edge late in life, he understood quantum mechanics as well as anyone through the 1930s, his philosophical objections to it were correct and should be taken more seriously, and his efforts effectively introduced the concept of quantum entanglement.
“I think that Einstein understood quantum mechanics as well as anyone, at least up through the 1930s. I think that his philosophical objections to it are correct.”
Dark energy evolving slowly across billions of years alarms particle physicists because it should be sped up by interactions, requiring a protective symmetry preventing interactions.
- Dark energy that pervades the universe and changes only over billions of years should set off alarm bells for particle physicists because particle physics deals in small length and time scales; any field evolving that slowly would, if it interacted with other particles, be sped up, so the only way to keep it slow and hard to detect is to impose a symmetry that protects it from interacting with other fields.
“you're telling me now I have a new field and its typical rate of change is once every billion years. That's just not natural.”
Objective reality exists; when a theory fits data, is simple, and makes internal sense, we should provisionally attribute reality to its important entities until a better theory appears.
- Objective reality exists, and when a theory fits the data, is simple, and makes sense in its own terms, we should attribute provisional reality to the entities that play an important role in that theory until a better theory comes along.
“If a theory fits the data and is very simple and makes sense in its own terms, then we say that the theory is right. That means that we should attribute some reality to the entities that play an important role in that theory, at least provisionally until we can come up with a better theory down the road.”
General relativity rests on spacetime as a unified four-dimensional object with geometry and curvature, where gravity is the curvature of spacetime itself rather than a Newtonian force.
- General relativity rests on the insight that spacetime is a unified four-dimensional thing that can have a geometry and be curved from place to place, and that gravity is precisely this curvature of spacetime rather than a force fitted into Newtonian mechanics.
“The final leap was to say gravity is the curvature of spacetime, and that statement is basically general relativity.”
Humans have an evolutionary bias to attribute intentionality and agency to anything acting human; Dennett's Intentional Stance captures that rationality about AI requires recognizing this bias.
- Human beings have a strong evolutionary bias to attribute intentionality, intelligence, and agency to anything that acts human—as people do even with Google Maps 'getting annoyed'—and Dennett's Intentional Stance captures that being rational about AI requires first recognizing this bias toward attributing hidden depth to surface-level human-acting systems.
“it's the most natural thing in the world for we human beings to attribute more intentionality to artificial things than are really there”
Emergence of complexity from simplicity is pre-paradigmatic and poorly understood; the key is likely information—how subsystems use information to survive, thrive, or reproduce.
- The emergence of complexity from simplicity is a fascinating but still pre-paradigmatic field where we haven't understood the basics; despite spectacular insights from people like Geoffrey West, the general principles remain to be found, and the key is likely information—subsystems of the universe figuring out how to use information to survive, thrive, or reproduce.
“I really do think that we haven't understood the basics yet. I think we're still kind of pre-paradigmatic.”
Life does not fight the second law of thermodynamics but surfs entropy increase by ingesting low-entropy fuel, maintaining non-equilibrium stability through entropy generation.
- Life does not fight against the second law of thermodynamics; rather, complex systems like us are surfers riding the wave of increasing entropy, maintaining stability dynamically by ingesting and burning low-entropy fuel to increase its entropy—we are non-equilibrium quasi-steady-state systems that rely on entropy increase to survive.
“Sometimes people think that complexity or life or whatever is fighting against the second law of thermodynamics ... That is precisely the wrong way to think about it. We are surfers riding the wave of increasing entropy.”
Time always ticks one second per second locally; it can differ between observers but feels the same to each observer's wristwatch, heartbeat, and neurological processes.
- Time always ticks by at one second per second locally; it can tick differently for different observers (as in the twin paradox where the traveler who takes the non-straight path returns younger), but as far as your own wristwatch, heartbeat, and neurological signals are concerned, everything happens at the same rate so time never feels funny.
“Time always ticks by at one second per second. That's all it can ever do. Time can tick by differently for different people”
The galaxy spans only tens of thousands of light years and billions of years old, requiring no relativistic speeds for self-reproducing probes to fill it.
- The galaxy is only some tens of thousands of light years across and billions of years old, so you don't need to move at a high fraction of the speed of light to fill it with self-reproducing probes.
“It's some tens of thousands of light years across and billions of years old. You don't need to move at a high fraction of the speed of light to fill the galaxy.”
Becoming a theoretical physicist creates pressure to drill infinitely deeper in a narrow area, when the harder skill is finding the best match between interests, abilities, and world importance.
- Once trained as a theoretical physicist you become the world's expert in an infinitesimally tiny area, creating an overwhelming temptation to just keep drilling deeper into what you started, when the harder and more valuable skill is stepping back to find the best match between your interests, abilities, and what is actually interesting to the world.
“There's an overwhelming temptation to just drill deep, just keep doing basically the thing that you started doing ... So you need to separately develop the capability of stepping back”
David Chalmers has persuasively argued that events in virtual reality and simulated realities should be treated as just as real as experienced reality.
- Philosopher David Chalmers has argued well that we should treat things that happen in virtual reality and simulated realities as just as real as the reality we experience.
“the philosopher, David Chalmers, has done what I consider to be a good job of arguing that we should treat things that happen in virtual reality and in simulated realities as just as real as the reality that we experience.”
All quantum mechanics versions require an arrow of time because measurement is irreversible, and many-worlds needs the same arrow as thermodynamics: simple, low-entropy initial state.
- All versions of quantum mechanics require an arrow of time because measurement is irreversible, and the kind of arrow of time many-worlds needs is exactly the one needed anyway for entropy and thermodynamics: a simple, low-entropy initial state.
“the kind of arrow of time you need in many-worlds is exactly the kind of arrow of time you need anyway for entropy and thermodynamics and so forth. You need a simple, low entropy initial state.”
Complexity starts low, rises, then falls to maximum entropy simplicity, so complexity peaks in the intermediate regime between minimum and maximum entropy.
- In an isolated system, entropy only goes up, but complexity starts low (few arrangements look like a simple state), rises, and then falls again at maximum entropy (everything smeared out looks simple), so complexity is highest in the intermediate regime between minimum and maximum entropy.
“entropy in this isolated system only goes up. That's the second law of thermodynamics. But complexity starts low, goes up, and then goes down again.”
The simplest and most likely correct answer to 'what is outside the universe' is nothing; the universe as the totality of everything has no outside and needs no larger world.
- It is legitimate to ask what is outside the universe, but one must accept that the simplest and most likely correct answer is that there is no such thing as outside the universe; if the universe is the totality of everything, it would not have an outside, and the universe is not obligated to obey the everyday rule that there is always a bigger world.
“you have to be comfortable with the possibility that the answer is there's no such thing as outside our universe. That's absolutely on the table. In fact, that is the simplest, most likely to be correct answer that we know of.”
Belief in God was reasonable but ultimately illusory given little knowledge; it is a substantive claim that proved wrong, replaceable by mechanistic understanding without loss.
- Belief in God was completely an illusion but a very reasonable one for people who understood little; it is a substantive claim about the world beyond verifiable predictions that turned out wrong, and had those people replaced it with a mechanistic universe they would have done just as well at understanding things.
“I think that was completely an illusion. I think it was a very, very reasonable illusion to be under.”
The question 'why is there something rather than nothing' lacks ordinary why-answers because the universe is not embedded in anything bigger, so it can simply exist.
- The question 'why is there something rather than nothing' does not have the kind of answer ordinary why-questions get, because typical why-questions are embedded in the universe and answered using its features, whereas the universe itself is not embedded in anything bigger or stronger and therefore can just be.
“the universe itself, as far as we know, is not embedded in anything bigger or stronger, and therefore it can just be.”
A star exhibits out-of-equilibrium dynamical complexity: macroscopically static yet burning fuel furiously to maintain stability, different from mere configurational complexity.
- A specific kind of complexity arises with stars: a star looks macroscopically stationary like a static sphere, but underneath the hood it is an out-of-equilibrium system burning fuel furiously to maintain that stability—distinct from configurational complexity since a star and a planet have similar configurations but the star only survives while it has fuel.
“it looks sort of macroscopically stationary, like a star, but underneath the hood, it's burning fuel to beat the band in order to maintain that stability.”
Neutrinos interact rarely with matter, requiring large dense transparent media for detection; Antarctic ice serves naturally as a neutrino detector for IceCube's photodetector arrays.
- Neutrinos punch right through the whole Earth and only rarely interact with matter, so a large transparent dense medium is needed to detect them; Antarctic ice is dense, transparent, and abundant, effectively making nature build a neutrino detector, which is how IceCube detects high-energy neutrinos with strings of 360-degree photodetectors looking for flashes in the ice.
“the nice thing about ice is it's transparent, so nature has built you a neutrino detector. That's what IceCube does.”
The principle of typicality claiming you are a typical observer is presumptuous and unjustified, so cosmological claims using it must be treated with extreme care.
- The cosmological move of reasoning by assuming you are a typical observer (the principle of typicality, mediocrity, or Copernican principle) is hilariously presumptuous, because asserting you are typical means asserting that typical observers are like you, which is completely unjustified—so such non-falsifiable cosmological claims must be treated with great care.
“by saying, 'I'm a typical observer in the universe,' you're saying, 'Typical observers in the universe are like me,' and that is completely unjustified by anything.”
Space-based solar collection is preferable to Earth-surface panels because the Sun provides fusion and orbital placement is mainly an engineering problem.
- Rather than covering Earth's surface with solar panels and data centers, it would be better to put solar collection in space, since we already have fusion in the form of the Sun and the only hard part—getting the energy down—is an engineering problem.
“Why would you waste the surface of the Earth with solar panels? Put them in space. We already have fusion. It's called the Sun.”
Panpsychism claims physics cannot capture subjective experience, requiring mind prior to matter; but consciousness is a real emergent non-fundamental phenomenon without changing physics.
- Panpsychism argues that physical descriptions of atoms and neurons can never capture first-person subjective experience (what it is like to see red), so something not contained in physics—'mind'—is needed, leading either to dualism or to putting mind prior to matter; but Carroll holds it is unnecessary because consciousness can be a real, emergent, non-fundamental phenomenon without changing the laws of physics.
“that first-person, inner subjective experience is simply not capturable by physics ... therefore is supposed to be, I need something that is not contained within physics to account for that, and I'm just going to call it mind.”
Most entropy in the universe today resides in black holes: supermassive black holes like Sagittarius A* contain more entropy than the entire observable universe once possessed.
- Most of the entropy in the universe today is in black holes: the whole observable universe had about 10^88 units of entropy before stars and black holes formed, yet the single supermassive black hole at the center of our galaxy has entropy of about 10^90—more than the entire universe used to have.
“The single black hole at the center of our galaxy has entropy, 10 to the 90. ... So most of the entropy in the universe today is in the form of black holes.”
The simulation hypothesis has no philosophical knockout objection but no real reason to take seriously; moreover, simulating realistic worlds is far harder than naively believed.
- There is no philosophical knockout objection to the simulation hypothesis, but also absolutely no reason to take it seriously; moreover, as a practical matter people will discover how much harder it is to simulate a realistic world than naively believed.
“I don't think that there's any philosophical knockout objection to the simulation hypothesis. I also think that there's absolutely no reason to take it seriously.”
As efficiency improves, people invent new things faster than efficiency gains, so environmental impact continues worsening despite efficiency improvements.
- Even as we get better at doing things efficiently, we invent new things we want to do faster than we improve efficiency, so we continue making environmental impact worse in various ways despite efficiency gains.
“we get better and better at doing things efficiently, but we invent more things we want to do faster than we get good at doing them efficiently. So we're continuing to make things worse in various ways.”
Cellular automata generate complexity from nothing but are irreversible with built-in time arrows, whereas real physics is reversible microscopically with macroscopic irreversibility from entropy.
- Cellular automata can generate wild complexity from near-nothing, but they are not how physics works because they are irreversible with a built-in arrow of time, whereas real physics conserves information and is reversible at the micro level with macroscopic irreversibility arising from entropy—so they don't help explain real-world complexity.
“It's also not how physics works at all because as we said, physics conserves information. ... These cellular automata do not, they're not reversible in any sense. You've built in an arrow of time”
The Nobel Prize likely benefits the world by highlighting good science but harms science itself by distorting motivations toward winning prizes.
- The Nobel Prize is probably a net good for the world because it brings attention to good science, but probably a net negative for science itself because it makes people want to win the Nobel Prize, distorting motivations.
“I think it's probably a net good for the world because it brings attention to good science. I think it's probably a net negative for science because it makes people want to win the Nobel Prize.”
Structural realism holds that theories' predictions point to real understanding regardless of changing fundamental ontology, so calling spacetime an illusion is wrong even if formulations are incomplete.
- Against the claim that spacetime is an illusion, structural realism holds that even though we don't know reality at the deepest level, the predictions a theory makes (e.g. when the sun rises) point to something true about our understanding regardless of changing fundamental ontology, so calling spacetime an illusion is simply wrong even if our best formulation is incomplete.
“that prediction, independent of what we thought the fundamental ontology was, the prediction pointed to something true about our understanding of reality. To call it an illusion is just wrong”
Complexity is not univalent: Kolmogorov, computational, and configurational complexity are genuinely different measures capturing different aspects.
- There is no unified measure of complexity because complexity is not univalent: Kolmogorov complexity measures how compressible a string of numbers is, computational complexity (e.g. traveling salesman, factoring) measures the difficulty of a problem, and configurational complexity measures the information needed to specify a system's configuration—these are genuinely different things.
“complexity is not a univalent thing. There's different ideas that go under the rubric of complexity.”
Consciousness and free will are emergent and non-fundamental but real like tables and chairs, unlike illusions that appear but lack causal efficacy.
- Carroll avoids calling consciousness an illusion: an illusion is something you perceive that isn't there and has no causal efficacy (like a desert oasis), whereas consciousness and free will are emergent and non-fundamental but just as real as tables and chairs, which are real even though made of atoms.
“I almost never use the word illusion. Illusion means that there's something that you think you're perceiving that is actually not there.”
Physics can help AI's compute efficiency problem because existing computers generate far more waste heat than brains, making computational efficiency improvement a major low-hanging fruit.
- Physics can dramatically help AI's compute problem because existing computers are hilariously less efficient and generate far more waste heat than brains; since the world of physics clearly permits these computations with much less waste heat, a major low-hanging fruit is improving computational energy efficiency rather than only scaling raw compute.
“the heat generation, the inefficiency, the waste of existing high-level computers is nowhere near the efficiency of our brains. It's hilariously worse”
Science can explain why people hold moral beliefs but cannot justify beliefs as right or wrong; science implements goals but never determines which goals are right.
- Understanding the function of the human mind can explain why people hold certain moral beliefs but will never justify those beliefs as right or wrong; science can help you implement and achieve your goals but cannot tell you which goals are right—if your goals are horrible, science is happy to help you achieve them.
“That will absolutely be helpful in explaining why certain people have certain moral beliefs, it won't justify those beliefs as right or wrong.”
The strongest skepticism of many-worlds is not empirical failure but that extrapolating Schrödinger beyond observation requires grand new philosophy about identity and probability.
- The strongest reason for skepticism about many-worlds is not that it fails to fit the data but that extrapolating the trusted Schrodinger equation beyond observation requires new philosophy—new ways of thinking about identity, probability, and prediction—which is doable but is a grand extrapolation rather than a simple step from what we already know.
“to take seriously the equation that we know is correct in other regimes requires new philosophy, requires a new way of thinking about identity, about probability, about prediction, a whole bunch of things.”
Einstein's equations are smarter than Einstein himself, predicting the Big Bang, gravitational waves, and black holes—none he anticipated.
- General relativity demonstrates a miraculous aspect of science: starting from clearly articulated assumptions, Einstein's equation is smarter than Einstein himself, because the equation predicted the Big Bang, gravitational waves, and black holes—none of which Einstein knew about.
“the equation is much smarter than Albert Einstein because Albert Einstein did not know about the Big Bang, he didn't know about gravitational waves, he didn't know about black holes but his equation did.”
Ideal podcast guests are mid-career: brilliant youth lack field mastery and wandering practice; overly practiced elders give canned answers; best conversations come from balanced expertise.
- The ideal podcast guest is mid-career: the most brilliant young people have less mastery over their whole field and are not practiced at wandering beyond their literal research, while certain older people give canned, rehearsed answers, so the best conversations come from someone with broad enough scope to range beyond their papers but not overly practiced.
“the most brilliant young people are not as practiced at wandering past their literal research ... Whereas, certain older people just have their pad answers and that's boring. So, you want somewhere in between”
Einstein could have figured out black holes; Schwarzschild derived the solution in 1917 just two years after general relativity, but physicists failed to recognize it until the 1950s.
- As smart as Einstein was, he never figured out black holes and could have; Schwarzschild derived the black hole solution to Einstein's equation in 1917, just two years after general relativity, yet no one recognized it for what it was until the 1950s—an example of physicists not being as clever as they should have been.
“as smart as Einstein was, he never figured out black holes, and he could have. It's embarrassing.”
Einstein's 1905 miracle year produced three groundbreaking papers on special relativity, atomic existence via Brownian motion, and photons via the photoelectric effect.
- Einstein's greatness is evidenced by his 1905 miracle year, in which he wrote three papers on wildly different subjects each worthy of fame: special relativity, Brownian motion (which proved the existence of atoms), and the photoelectric effect (which invented photons by showing light should be thought of as particles as well as waves).
“in 1905, his famous miracle year, he writes three different papers on three wildly different subjects, all of which would make you famous just for writing that one paper.”
Large language models are trained to predict and sound human through optimization wildly different from human evolution, so judging them against humans misses understanding what they actually are.
- Large language models are trained to predict text and are optimized to sound human, which is a wildly different process and objective than how humans came to be, so judging them against human beings misses the chance to understand what they actually are—even though there is no in-principle obstacle to building a machine indistinguishable from a human in cognition.
“What a large language model is trying to do is to predict text. That's what it does.”
Carroll's writing process involves thinking silently for long periods before writing near-final text, unlike typical heavy revision practices.
- Carroll's writing process differs from typical free-writing: rather than drafting and heavily revising like his wife, he sits silently thinking for a very long time and then writes what is almost the final draft, so most of the work happens in his brain before it reaches the page.
“I will just sit there silently thinking for a very long time and then I'll write what is almost the final draft.”
People sort into a grid of agreement and respect levels including 'embarrassing allies'; productive discourse should explain agreement and disagreement rather than psychologize motives.
- People can be sorted on a two-by-two grid of whether you agree or disagree with them and whether you respect them or not, and all four quadrants are populated—including the 'embarrassing allies' quadrant of people who agree with you but whom you have no respect for—so productive discourse should focus on explaining agreement and disagreement rather than liking.
“I had a two by two grid and it's are you someone I agree with or disagree with, are you someone who I respect or don't and all four quadrants are very populated.”
In spacetime, proper time along your trajectory is like distance along a space curve but inverted: longest proper time between events is a straight line, opposite space's shortest-path rule.
- In spacetime, the time you measure along your trajectory is analogous to distance along a curve in space, but inverted: whereas the shortest distance between two points in space is a straight line, the longest proper time between two events is a straight line, so zigzagging through spacetime makes you take less time from point A to point B.
“the longest time between two points is a straight line. If you zig back and forth in spacetime, you take less and less time to go from point A to point B.”
Accusations of arrogance usually mean disagreement; forward conversation requires specifying substantive disagreement rather than psychologizing motives or insulting.
- When people accuse you of being arrogant, almost always all they mean is that they disagree; to move a conversation forward you should specify the substantive disagreement rather than psychologize the other person by attributing motives, because an actual response to their argument is far more productive than insults.
“when people react against you by saying you are being arrogant about this, 99.999% of the time, all they mean is I disagree.”
General relativity predicts a past singularity of infinite curvature signaling theory breakdown; there may have been a first moment of time or preceding time evolved into our Big Bang.
- General relativity unambiguously predicts a past singularity of infinite curvature, which really signals the theory breaking down; it is genuinely possible there was a first moment of time before which no moments existed (the Big Bang), but equally possible there was time before that evolved into our hot Big Bang—we have no idea why the Big Bang happened.
“It is completely possible there was, in some sense, a moment of time before which there were no other moments. And that would be the Big Bang.”
Probes reading symbolic Othello moves from deep learning networks found evidence the network invented internal board representations, raising questions about whether optimal action requires human-like processing.
- Experiments feeding a deep learning network symbolic Othello moves found that a probe network could read off whether positions were legal or pieces were a given color substantially better than chance, providing tentative evidence that the network invented an internal representation of the Othello board—an open question being whether the most efficient way to act human is the same way humans do it.
“the little probe network can ask, 'Would this be legitimate or is this token white or black?' ... it has invented the Othello board. And it found that the probe got the right answer, not 100% of the time, but more than by chance, substantially more than by chance.”
Human imagination may arise from evolutionary shift when fish moved to land: underwater requires sub-second sight-reaction decisions, while land's far vision enabled 'what-if' reasoning before acting.
- There is an argument that what separates humans from other species is our ability to imagine counterfactual hypothetical futures, a capacity tied to a cognitive shift when fish moved onto land: underwater, short light-attenuation forces sub-second react-on-sight decisions, but on land you can see far and even see stars, opening a mode of reasoning where you contemplate 'what if I did this' before acting—the birth of imagination.
“There's an argument to be made that literally what separates human beings from other species on Earth is our ability to imagine counterfactual hypothetical futures.”
Focusing on AGI mistakes artificial intelligence as humanity-bound; actual agents excel at some things and fail at others, requiring appreciation of actual capabilities over anthropomorphism.
- The mistake in focusing on AGI is treating artificial intelligence as if it were on a path toward human-like intelligence; instead, an artificial agent may be far better than humans at some things and far worse at others, so we should appreciate its actual capabilities rather than anthropomorphizing it, which is both more accurate and better for safety.
“artificial intelligence is different than human intelligence, and so the mistake that is being made by focusing on AGI ... rather than trying to ask, how close is it to being a human-like intelligent, we should appreciate it for what its capabilities are”
Given nuclear weapons and bioweapons, civilization's survival is surprisingly fortunate, and confidence we won't go too far is exactly what one says before disaster.
- Given the number of nuclear warheads and the existence of bioweapons, it is genuinely surprising that humanity has not yet caused enormous devastation or destroyed civilization, and Carroll cautions that confidence we won't go too far is exactly what one would say right before going too far.
“Between nuclear and bioweapons, it is a little bit surprising that we haven't caused enormous devastation.”
Poetic naturalism holds only nature exists, but many valid ways describe it; any real and causally efficacious description gets reality attributed, including normative and aesthetic vocabularies.
- Poetic naturalism holds that the natural world is all that exists (no supernatural), but that there are many valid ways of talking about it—as long as a description latches onto something real and causally efficacious we attribute reality to it—including not just descriptive vocabularies but normative, prescriptive, and aesthetic ones that, while subjective, are not arbitrary.
“there are many ways of talking about the natural world and, as long as those ways latch onto something real and causally efficacious about the functioning of the world, then we attribute some reality and truth to them.”