What this covers

Mark Zuckerberg sits down with Patrick Collison and Tyler Cowen to examine a troubling paradox: despite exponentially larger investments in scientific research, education, and infrastructure over the past century, the measurable returns have stalled or flatlined. The conversation moves across economics, science funding, institutional design, and demography to argue that progress is not a fixed rate but a choice—one that depends heavily on how societies structure their incentives, institutions, and ability to experiment. Rather than assuming innovation is intrinsically harder, the three explore the possibility that existing institutions have simply become sclerotic, and that deliberate effort to understand and revive the conditions for progress could unlock substantial gains.

The discussion anchors on concrete examples: microscopic measures like PhDs and research dollars have grown 50 to 100-fold since mid-century, yet crop yields, life expectancy gains, and semiconductor advances show diminishing or constant returns. Infrastructure offers the sharpest case—New York built 23 subway stations in under five years around 1900; a century later it took 17 years to build three Second Avenue stations. Cost Disease appears across healthcare, education, and housing, driven by bureaucratic bloat, restricted entry, and distorted incentives rather than physics. The speakers examine why Diminishing Returns to Inputs persist despite knowing that alternatives work: Singapore achieves higher life expectancy on 4% of GDP healthcare spending; France caps treatments per condition and negotiates drug prices more effectively; housing in Japan stays affordable through regulatory design. They trace obstacles to Veto Power Proliferation, aging patterns in research funding, and institutional stagnation—tenure and university rankings have barely shifted since 1920—before turning to what deliberate experimentation and historically-informed "progress studies" might reveal about rekindling momentum.

Sharpest takeaway

Collison and Cowen argue that humanity's rate of progress is not fixed and appears to be slowing despite massively increased inputs, so we urgently need a dedicated, experimental, historically-informed 'progress studies' effort to understand how progress happens and how institutions can generate it more effectively.

  • Across applied domains (life expectancy per dollar, infrastructure, crop yields), exponentially increasing inputs yield at best constant returns, which is unsustainable and points to institutional rather than intrinsic causes
  • Existing relevant research exists but is sparse and under-attended; the central questions (NIH effectiveness, cost disease) lack definitive answers
  • Experimentation at the institutional level (universities, science funding, philanthropy) has atrophied, and reviving it could unlock large gains

The argument · threads4 threads · 47 claims
0.85

Institutional and sociological barriers—not legal or technical ones—now obstruct progress across housing, infrastructure, science, and higher education.

6 pointscentrality 5/5
  • Hsieh and Moretti's updated model shows that if Bay Area and New York zoning restrictions between 1964 and 2009 had instead allowed much greater housing supply elasticity, average US income would be about $3,700 higher per person—not just for residents of those areas but across the whole country—implying a huge effect from housing policy that should be studied and acted upon.

    “if you look at the zoning restrictions that existed in the Bay Area and New York between 1964 and 2009, and you imagine a counter-factual world where there was much more supply elasticity... average U.S. income would be, in their model, $3,700 higher per person”

  • The entry costs to forming a new university are extremely high, but not through formal legal tolls like zoning; instead the barriers are sociological and institutional—accreditation dynamics and practical realities—making it apparently almost impossibly difficult to create a successful new university today, which is one reason answering the cost-disease question is so important and surprisingly unresolved.

    “the entry costs of forming a new university are really high, but that’s not because there’s a kind of formal toll you have to pay... it’s apparently almost impossibly difficult to create a successful new university today”

  • In 1980 the NIH spent 12 times more dollars on researchers under 40 than over 50, predominantly funding younger people, but today it spends 5 times more on those over 50 than under 40—an inversion toward a 'gerontocracy' whose effect on scientific productivity is unknown but is an important policy question.

    “in 1980 the NIH spent 12x more dollars on researchers under 40 than researchers over 50... Today they spend 5x more dollars on people over 50 than under 40”

  • Larry Summers' idea of the 'promiscuous distribution of the veto power' explains why it has become so much harder to get things done: as the ability to block projects is spread across more and more actors, large undertakings stall, and this dynamic likely extends beyond infrastructure to other progress-generating mechanisms in society.

    “Larry Summers talks about the idea of the promiscuous distribution of the veto power and how much harder it is to get things done”

  • Infrastructure productivity has fallen dramatically for non-physical, institutional reasons: New York built 23 subway stations in 4.7 years for just over a billion (2019) dollars around 1900 but only 3 Second Avenue stations in 17 years for $4.5 billion by 2019 (a roughly 40x productivity decline); the Golden Gate and Bay Bridges plus a man-made island were built in under four years in the 1930s; and France opened its first TGV line in five years while California's high-speed rail is projected to take 25-plus years—proving the slowdown is institutional and sociological, not because atoms got heavier.

    “When New York decided to build the Second Avenue subway in the year 2000, 17 years later they opened three stations and they spent $4.5 billion doing so. And so our productivity in subway construction has, at least in New York, decreased by a factor of 40”

  • Cost-disease-type problems are hard to diagnose because the cause can be diffuse: for example, if scientists' administrative burden had increased by two-thirds over 40 years it would manifest only as longer approval times, longer forms, and more interruptions—hard to observe and hard to causally attribute—so specifically identifying the causal pathways of slowdown is genuinely tricky across many fields.

    “if it were the case that the administrative burden on scientists had increased by say two-thirds over the last 40 years... that might be quite difficult to observe because it can come in the form of, well, it takes twice as long on average for things to be approved and the forms are longer and you’re interrupted more.”

0.75

Progress accelerated sharply around 1700-1750 and remained strong through mid-century but has since decelerated despite vastly larger inputs.

4 pointscentrality 5/5
  • By any major metric—average income, life expectancy, infant mortality—the rate of human progress was constant or improving only very slowly between 0 and roughly 1700-1800, and then something changed around 1700-1750 (Industrial Revolution, Enlightenment, modern science) after which incomes, life expectancy, fundamental knowledge, and new technologies all began improving together and diffusing globally.

    “the rate of progress has not been constant... between 0 and 1700, 1800, thereabouts... it was either constant or only improving at a very slow rate... And then something happened, something changed around sort of 1700, 1750”

  • US productivity growth was roughly 1.9% a year mid-century (about 1920-1970), but most economists now think it is much lower, around 0.4% a year, meaning that by economic measures we are generating progress far more slowly than we used to.

    “if you look at the U.S., productivity growth mid-century or say between, you know, 1920, 1970, was maybe about 1.9% a year. Now most economists think it’s much lower, maybe around sort of .4% a year”

  • Microscopic measures of scientific input—number of PhDs, papers published, dollars of funding—have grown by factors of roughly 50 to 100 (orders of magnitude) between the first and second halves of the twentieth century, yet for almost every applied output measure (life expectancy, semiconductors, crop yields) we get at best constant returns; exponentially increasing inputs with constant outputs is by definition unsustainable, and the cause may be institutional and sociological rather than progress simply getting intrinsically harder.

    “for almost every conceivable applied measure we seem to be getting at best constant returns. But that’s really bad because we see exponentially increasing inputs and constant return outputs. That is almost by definition not a process that we can sustain”

  • The number of people in extreme poverty has declined by more than a billion since Collison was born, and global life expectancy has risen about six years while infant mortality has fallen more than 50%, constituting a moral imperative not to lose sight of the value of this progress.

    “the number of people in extreme poverty has declined by more than a billion people since I was born”

0.66

Historical examples and empirical studies demonstrate that organizational design, funding structures, and management practices directly enable progress.

5 pointscentrality 4/5
  • Multiple studies—a Stanford randomized trial in India, Michela Giorcelli's study of Italian firms showing benefits 15 years after management training with natural randomization, and a randomized trial of about 600 firms in Mexico—suggest that simply teaching better management practices makes firms employ more people, pay higher wages, and become more successful, which if true is amazing low-hanging fruit warranting much more investment and study of which practices generalize.

    “there’s a really neat one that came out last year from Michela Giorcelli looking at firms in Italy and showing that over 15 years after a management training program with some natural randomization, that those firms were employing more people, paying more wages, being more successful”

  • Seven of the world's top 25 universities today were founded in a single 30-year period between 1861 and 1891 by people who were deliberately reform-minded—inspired by the German research university model and the possibilities of the U.S.—who chose to try something different, empirically underscoring the value of institutional experimentation and commitment.

    “Seven of the top 25 are American universities that were started in a single 30-year period between 1861 and 1891... they were very deliberately and specifically reform-minded.”

  • History shows extraordinary creative clusters—Vienna 1880-1940 (Klimt, Mahler, Mach, von Mises, Hayek, Freud, Wittgenstein) and Edinburgh in the Scottish Enlightenment (Smith, Hume, modern geology, literature) despite being tiny—where many of the productive people knew and inspired each other across disciplines, showing that something locally excellent (hard to pin down but real) drove disproportionate output relative to comparable places like Dublin in 1780.

    “Look at Vienna, 1880 to 1940... a lot of these people knew each other and they were inspired by each other... clearly there was something excellent in Edinburgh in 1780 that was not there in Dublin in 1780”

  • Pierre Azoulay's study found that among ostensibly identical scientists, those who received longer-term, open-ended HHMI grants were about twice as likely to produce a top-1%-by-citation paper as those who did not, which is suggestive—while controlling for selection effects—that open-ended funding structures better enable scientific discovery.

    “he concluded they’re about twice as likely to produce a top 1% paper by citation count”

  • The historical examples of fast, ambitious projects show how effectively humanity can accomplish things when organized the right way, so when possibilities are unlocked and efficacy is enabled we can do great things.

    “it’s just important to understand how effectively we as a species can do things when we’re organized the right way. Humanity is pretty amazing.”

0.53

Defining progress rigorously and investigating its major drivers remain surprisingly under-studied despite their importance to policy and institutional reform.

5 pointscentrality 3/5
  • In 1980 about 40% of Americans moved somewhere else when they took a new job, but last year only about one in ten did, a decline driven largely by enormously increased costs of movement as housing costs rose in the most productive regions over recent decades.

    “if you look at the USA in 1980, 40% of people moved somewhere else when they took a new job... If you look last year, about one in ten people moved when they took a new job”

  • As far as can be told there are no books assessing how well the NIH is working—which parts work better or worse, or whether the old NIH was better than the new one—even though this is among the most important questions; the gap shows that major swathes of progress-relevant questions are surprisingly under-investigated despite plentiful good research.

    “as far as I can tell, there are no books assessing how well the NIH is working... I do have a strong view on the importance of knowing.”

  • GDP or GDP per capita is a pretty good first-approximation measure of progress and correlates strongly with many things we care about, but they are not the same thing, so an important open question is how to define progress and what its better and worse kinds are.

    “I think GDP or GDP per capita is a pretty good first approximation measure of it and it correlates strongly with many of the things we care about. But they’re definitely not the same thing.”

  • For the most important questions—how to organize science, which policies generate growth, how to diffuse innovation—definitive data does not exist, so existing findings should be treated as food for hypothesis-generation rather than answers; the powerful approach combines marshaling that evidence with a willingness to experiment and to be wrong, as exemplified by Warren Weaver funding Norman Borlaug: informed by empirical realities yet willing to place a bold bet without strong prior conviction.

    “I think of the data, such that it exists, and the exiting findings as food for hypothesis-generation.”

  • Even if technology like telepresence solves the agglomeration imperative of cities, the socio-institutional dynamics that have made cities less effective probably also generalize to other domains, so we would still suffer those same costs elsewhere—meaning technological fixes don't automatically escape the underlying institutional problems.

    “even if technology solved that, I guess my worry would be that the socio-institutional dynamics that have ruined cities or made them less effective or whatever, probably also generalized and applied to other domains. And so we’re going to suffer the costs of those same phenomena elsewhere.”