What this covers

Michael Nielsen and Tyler Cowen range across science, technology, and institutions in this wide-ranging conversation, with Nielsen developing a sustained argument about what enables progress at civilization's frontier. He contends that breakthrough work is fundamentally idiosyncratic—resistant to standardization or scaling—which means the people who identify talent and build collaborations must themselves think in unconventional ways. He illustrates this through case studies: Paul Ginsparg's success with the physics preprint server came not from universal ambition but from narrowness and leverage over a few high-status figures. Nielsen also explores how staying creative across a career requires proximity to younger minds and network effects rather than accumulated authority, and he argues that active vs passive openness—shared code and notebooks—matters far more for scientific progress than freely available journal articles alone.

The conversation then widens to examine how technology progressively embeds governance into infrastructure itself. Drawing on Neil Postman and Lawrence Lessig's Code is Law framework, Nielsen warns that control over cognition and action is shifting from individuals and populations toward centers of capital as more of human capability moves into devices and systems. He touches on quantum computing as a genuinely open problem that might overturn existing theory, on why mathematics maintains healthier intellectual diversity than theoretical physics, and on why institutional structures like journal bundling and economies of scale narratives actually damage individual researchers' leverage. The pair also debate more abstract territory: whether the question "why is there something rather than nothing" is meaningful at all, whether quantum mechanics underlies consciousness, and what scale of off-world civilization would be viable for long-term human survival.

Sharpest takeaway

Nielsen argues that progress at the frontiers of science and civilization depends less on status, capital, or institutions than on idiosyncratic individuals, diversity of thought, sustained attention, and proximity to young, active minds — while warning that technology increasingly embeds governance and creates civilizational vulnerabilities.

  • Frontier work is idiosyncratic and resists commoditization, so talent identifiers and collaborators must themselves be idiosyncratic
  • Staying creative across a career depends on having younger mentors and positive network effects, not accumulated authority
  • Technology progressively moves governance and cognition into devices, shifting power away from populations and individual researchers toward capital and infrastructure

The argument · threads10 threads · 39 claims
0.75

New fields and creative breakthroughs favor people unburdened by sunk costs and status concerns.

4 pointscentrality 5/5
  • New fields of science tend to be dominated by people in their 20s for essentially sunk-cost reasons: people who have no sunk costs in existing paradigms are freer to go into nascent fields, as seen in the early days of quantum computing and open science.

    “those tend to be dominated essentially, for almost sunk-cost reasons — people who don’t have any sunk costs tend to be younger. They go into these fields.”

  • If you are searching for comparative advantage in doing creative work, you want to know where status is — but mostly so you can avoid it, echoing Peter Thiel's advice to 'be short status,' because the people who remake fundamental understanding are very good at ignoring status.

    “If you’re searching for comparative advantage in doing creative work, you want to know where status is, but mostly so you can avoid it.”

  • Unfortunately for creative work there is often an anticorrelation between how valuable what you are doing is and what you are being paid; if an institution is keen to pay you to do what you're doing, you should consider whether you're in the right line of business — a deliberately anti-economist heuristic that informed Nielsen's decision to leave AI for other work.

    “I tend to think, unfortunately for creative work, there’s an anticorrelation between how valuable what you’re doing is and what you’re being paid, often.”

  • There is no tension between making things legible and valuing illegibility: as you constantly expand the legible, a penumbra of illegibility surrounds and moves with it, giving you access to new spaces — so making the world's culture and people more legible to yourself expands what you can see at the edge.

    “You’re constantly expanding the legible, and when you do that, there’s this penumbra of illegibility that surrounds it. That moves, but it gives you access to those other spaces.”

0.75

Intellectual diversity and disagreement matter far more than consensus for driving scientific progress.

4 pointscentrality 5/5
  • Mathematics is a healthier intellectual culture than theoretical physics because each mathematician tends to be unique with their own path and beliefs, whereas theoretical physicists are more monotone and can sum themselves up in a few words to colleagues — and intellectual diversity, not consensus, is what matters for progress.

    “I find mathematicians — when I talk to them, it’s such a healthy culture because each mathematician is really, well, a lot of them are very unique... Theoretical physics often seems just a little bit more monotone. They can sum themselves up in a few words when they’re talking to their professional colleagues, and that’s not so healthy.”

  • Examining older physicists who aged well versus those who grew complacent, having younger mentors was the key differentiating factor; through a network effect, surrounding yourself with active 25-to-28-year-olds whom you regard as mentors (not as your students) lets you partake of positive network effects and stay at the edge.

    “having younger mentors was really the key thing... it’s almost a network effect. Basically, if there is some slight downhill slide, and most of your friends are not quite at the edge anymore, that’s going to infect you. If you still have mentors who are 25, 28, extremely active... you get to partake of the positive network effects.”

  • Beyond shared traits of openness, curiosity, prosociality, ambition, and imagination, Nielsen almost always selects collaborators who have at least one very strong skill that he does not have, treating that as sufficient diversity.

    “I will select for somebody who has at least one very strong skill which I do not have. That’s enough diversity from my point of view.”

  • Although expertise research (Herb Simon, Anders Ericsson) claims experts have acquired tens of thousands of chunks of information, the causal relationship is murky; Nielsen suspects that deep expertise and broad knowledge are downstream of something else — determination and curiosity — rather than memory being the primary cause.

    “are you an expert because you know those things? Or is it really downstream of something else?... I suspect actually, it’s downstream of something else — your determination, curiosity, something like that.”

0.71

Open, shareable practices like preprints and Jupyter Notebooks accelerate scientific progress far more than institutional publishing.

6 pointscentrality 4/5
  • Whether a field adopts preprints is a contingent cultural matter more like fashion than a function of competitiveness — physicists and biologists each invoked their field's greater competitiveness to explain opposite behaviors, showing the explanation is post-hoc rationalization rather than the real cause.

    “The biologists would say, “Biology is so much more competitive than physics that we can’t possibly bear to share our results too early.” The physicists would say, “Physics is so much more competitive than biology that we have to share them as rapidly as possible.””

  • Putting his neural nets textbook online for free generated far greater impact and ultimately benefited Nielsen more financially — through reputation, opportunities, and jobs — than any amount of royalties would have, so for authors paid royalties make little sense even though publishers' incentives differ.

    “I wrote this neural nets textbook, which I put online for free, and that’s made a really large difference to the impact which it’s had. Even if I just think purely financially... the greater impact has actually benefited me much more financially than any amount of royalties ever would.”

  • 'Openness' per se is too undefined to be useful — it is always relative to platform, institutions, and norms — but specific active forms like openly shared Jupyter Notebooks in machine learning have genuinely driven a lot of progress, far more than merely making journal articles freely available, and as norms change open practices could speed science by more than 2x.

    “the culture around Jupyter Notebooks in machine learning, I think having those very openly available and widely available really has driven a lot of progress... It’s not the same as making your journal article openly available. It’s a much more active kind of a material.”

  • Paul Ginsparg made the physics preprint server succeed by going very narrow rather than trying to solve the problem across all fields, twisting the arms of a few high-status physicists like Andy Strominger and Ed Witten to contribute their best papers, then agglomerating other communities outward from that tiny core.

    “He went very narrow. He didn’t try and solve the problem all across all the fields. He went and twisted the arms, at some level, of some very high-status, high-profile physicists... then you can agglomerate. You can start to attach other communities.”

  • Librarians, who understand as well as anyone that their proxy is imperfect, default to using impact factor as the differentiator for which journals to subscribe to because they have nothing better to do, illustrating reliance on a flawed metric in the absence of alternatives.

    “they will use impact factor. That’s the differentiator that they tend to use... They’re just using a very imperfect proxy. They understand, as well as anybody, that their proxy is imperfect, but they don’t have anything better to do.”

  • The widely believed theory that scaling is very important is a capital-friendly story that shifts negotiating power away from individual researchers toward centers of capital, while it remains 'just a story'; combined with big companies publishing fewer papers, this damages individual researchers' ability to build public brands and capture value.

    “Scaling is a very capital-friendly story. It actually moves some of the negotiating power from individual researchers, I think, to the centers of capital, but it is just a story.”

0.56

Governance and control are increasingly embedded directly into technology rather than remaining in human hands.

3 pointscentrality 4/5
  • Following a Neil Postman and Lessig 'Code is Law' view, more and more governance infrastructure is being built directly into technology, moving control out of the hands of the population and into the technology itself — the story of the last 100 years and likely the next 100.

    “You just keep building more and more governance infrastructure into the technology, and you’re moving it out of the hands of the population and into the technology. That seems to be, certainly, the story of the last 100 years, and very likely the story of the next 100 years.”

  • Surveillance carries negative connotations but is necessary because supplying justice depends on understanding what has occurred; however, an optimal system needs a mix of surveillance and opaqueness — latitude to misbehave and break laws — to keep the system stable and limit the abuses of power by the powerful, ideally enforced through Madisonian conflict between powerful institutions.

    “our ability to supply justice is dependent upon having a good understanding of what has occurred in the present and in the past.”

  • Bad journals persist because the person getting the utility (the researcher) is not the person making the buying decision (the library), which is always bad, and because the rise of internet-era economies of scale led libraries to subscribe to giant bundles rather than individual journals, allowing crummy titles to ride along.

    “One is that libraries pay, not individuals, usually, for subscriptions. The person getting the utility is not the same as the person making the buying decision. That’s always bad... Libraries don’t subscribe to individual journals for the most part. They subscribe to all these giant bundles.”

0.56

Identifying and connecting emerging talent requires idiosyncratic judgment that cannot be commoditized or systematized.

3 pointscentrality 4/5
  • The best talent identifiers, like Stewart Brand, are abstractly similar to one another but identify very different talent; they function as human marketplaces or crossroads, connecting people to opportunity, and such people are very rare — Nielsen has met only five to ten in his life.

    “they’ve become human marketplaces as well. Actually, they’re at a crossroads. They’re connecting people to opportunity. That’s a very special type of person. I’ve maybe met five to ten people in my life who seem like that.”

  • Great talent identifiers are idiosyncratic — like Michelin chefs — because the boundaries of knowledge at any given time are themselves idiosyncratic almost by definition, not yet commoditized, so identifying people who can expand those boundaries requires having some sense of that edge.

    “it’s just because the boundaries of knowledge at any given time tend to be idiosyncratic almost by definition. They haven’t been commoditized yet.”

  • Nielsen became enthusiastic about Emergent Ventures empirically after meeting grantees doing very worthwhile things that have little or no institutional chance of support, and he reads Cowen's willingness to fund people with anti-libertarian ideas as testing whether they genuinely care about their idea rather than ideological alignment.

    “seeing so many people who are doing very worthwhile things which have no or very little institutional chance of support, being amplified — that I care about a great deal, and it seems like EV is one of the places that is doing that the best.”

0.53

Pushing physical systems into new regimes through quantum computing may reveal limits and flaws in fundamental theory.

5 pointscentrality 3/5
  • Progress in fundamental physics has stalled in part because the fundamental theories have been too successful for the last 50 years; pushing quantum-coherent systems into new regimes (as in quantum computing) creates a chance theories break down, which would be productive because we would learn a lot.

    “The problem in some ways in physics has been that the fundamental theories have been just too successful for the last 50 years.”

  • The most interesting outcomes of quantum computing would be discovering that quantum mechanics is wrong, or discovering new materials, because quantum computers could perform vast numbers of accurate quantum-scale simulations far faster than current methods, which are terrible and often produce wrong results.

    “It’s very hard to do simulations of stuff down at the quantum scale. The ways that we have are pretty terrible and often produce wrong results. The fact that we may actually have a very high-throughput way of doing lots and lots of simulations which give correct results — it’s being able to do a thousand times as many experiments as before.”

  • Quantum computing is an extremely difficult, definitively unsolved problem, but the existence of order-100-qubit systems that can apparently be manipulated at will suggests it is going to happen eventually if we just wait — though what it will mean is unknown.

    “the fact that there’s order-100-qubit systems, which you can apparently manipulate as you will, suggests to me we just wait. It’s going to happen. We don’t know what it will mean.”

  • People who know the field point at ASML as surprisingly hard to duplicate — not the lithography itself but doing the lithography at scale — making it a plausible chokepoint where monopoly power in a strong-AI future concentrates, an amusing return to Dutch Renaissance strength.

    “they all point at ASML as having been surprisingly hard to duplicate. It’s not so much do the lithography, but do the lithography at scale, which seems to be very, very hard.”

  • AI is very likely to drive a lot of scientific progress over the next few years because we are moving much of our cognition — and eventually our actuators, the way we operate in the world — out into devices, where it becomes much more mutable and improvable.

    “I think it’s very likely to drive a lot of scientific progress over the next few years. That’s just a case of, we’re moving so much of our cognition, and eventually, also the actuators, the way we operate in the world, out into these devices, where all of a sudden it becomes much more mutable and hopefully improvable.”

0.34

Long-run economic growth will eventually make establishing permanent human civilization in space economically viable.

4 pointscentrality 2/5
  • Establishing a civilization off Earth is utterly crucial for the long-run survival of civilization; the right scale is roughly a million people — not 20 — which carries enough civilizational infrastructure to be viable even if not fully self-sustaining, and continued economic growth will steadily lower the relative cost of getting there.

    “I think getting off planet Earth and establishing a civilization elsewhere is very, very important... Very hard for economic reasons but utterly crucial.”

  • Continued long-run economic growth will steadily lower the relative cost of going to Mars or establishing a permanent human presence in space, until at some point it is no longer that difficult, even though it is currently very expensive given the return.

    “If we continue to have economic growth, the relative cost is just going to keep going down. At some point, it’s actually not going to be that difficult.”

  • Olaf Stapledon's central analytical mistake in his future histories was assuming away collective-action problems and the difficulty of supplying public goods — dissolving them narratively without a mechanism, even though such problems are precisely the interesting ones he should have engaged.

    “I think he didn’t really believe in them, or actually understand just how difficult they are to solve, how difficult it is to supply public goods and these kinds of things. He always does away with it, narratively, and it’s assumed away without really a mechanism being given.”

  • An aircraft carrier like the Midway represents an almost perfect floating civilization because being on a boat removes the conditions for empire-building and organizational bloat — there are real reasons to trim fat — producing rare unity of purpose, clarity of command, and genuine shared belief in a high purpose among its thousands of crew.

    “They don’t suffer like large organizations that have all kinds of bloat and all kinds of problems. So many of those problems are gone away there, partially because when you’re on a boat, it’s not so easy to empire build, like there are real reasons to trim fat.”

0.29

Humans perceive only a tiny fraction of the universe, complicating explanations of why cosmic structure feels beautiful to us.

3 pointscentrality 2/5
  • We seem to be programmed to recognize and find attractive, instinctively, novelty which is associated to structure, which partially explains why large-scale cosmic structures like spiral galaxies appear beautiful to us — though the same explanation fails to explain why cosmic sounds are not beautiful.

    “we do seem to be programmed to recognize and find attractive (instinctively) novelty, which is associated to structure somehow. So, we look and we see spiral galaxies or things like that.”

  • Humans see in such a tiny band of wavelengths — not into the infrared, radio, ultraviolet, or X-ray — that we are really seeing almost nothing of the universe, which complicates evolutionary explanations of why cosmic structure appears beautiful to us.

    “We see in such a tiny band of wavelengths. We’re really not seeing almost anything. We’re not seeing into the infrared and the radio. We’re not seeing into the ultraviolet and the X-ray.”

  • The James Webb Space Telescope, costing about $10 billion, is arguably a machine for producing beautiful images of the universe — one of the most important and most expensive image factories ever made.

    “The James Webb Space Telescope was about, I think, $10 billion — and is arguably a machine for producing that kind of image. It’s got to be one of the most important image factories and most expensive image factories ever made.”

0.29

Paternalistic institutions deny individuals basic agency, a concern that has shaped Nielsen's life and values.

3 pointscentrality 2/5
  • Drawing on Ivan Illich's Deschooling Society, Nielsen internalized that schools do not treat children as fully human and deny them the most basic claims of agency, which made the question of how paternalistic institutions should be toward individuals central to his entire life.

    “In Deschooling Society, he really makes the point that, in fact, schools do not treat children as human to some extent at all. It denies them the most basic claims of agency.”

  • Nielsen describes himself as extremely disagreeable (though polite and kind), and the younger mentors he has attracted are all people who enjoy disagreement — they engage when offered another way of looking at things, whereas people who feel insulted or threatened by disagreement make poor collaborators.

    “if you look at all of the younger mentors I’ve had in the last, say, seven or eight years, they’re all people who enjoy disagreement... Some people get insulted, or they get threatened, or they get annoyed when you do that, and those people — they’re not going to be good collaborators.”

  • The deeper value of practices like spaced repetition, taking photos, and taking notes is not just better recall but that they are general-purpose strategies that cause you to pay attention to the world in different ways — which is incredibly valuable — and this is the same reason Nielsen agrees to do podcasts that force novel attention.

    “Any general-purpose strategy you have which will cause you to pay attention to the world is incredibly valuable, so I collect things like that. Why did I say yes to coming on the podcast? A huge part of it is because I know it’s going to make me pay attention in different ways.”

0.15

Fundamental cosmology has made almost no progress on deep questions despite centuries of inquiry.

3 pointscentrality 1/5
  • Nielsen's probability for the existence of a mystical, prior-to-physics God has not changed since he was seven, because the foundational cosmological questions — why is there anything at all — have seen essentially no progress in the 40-odd years since, which he finds frustrating given how much else he has learned.

    “As far as I can tell, we haven’t made any progress on those things in the 40-odd years since. It’s frustrating, actually, that that’s the case.”

  • Nuclear weapons are terrible but not directly civilization-threatening; the more pressing concern is whether cost and expertise to build a city-destroying weapon fall to the hundreds-of-thousands-of-dollars level, and Nielsen is actually more concerned about biosafety and computer security threats.

    “The issue is, nuclear weapons are terrible, but they’re not civilization threatening directly.”

  • While much structure in the brain is quantum mechanical (for instance, the reason atoms are stable involves quantum mechanics), Roger Penrose's idea that the brain is a quantum computer in a way that affects consciousness is probably not correct.

    “It’s certainly possible that there’s some very interesting structure in there that is quantum mechanical in some really interesting way... but it probably doesn’t affect anything about consciousness or anything like that. I would be very surprised if that were the case.”