What this covers

Patrick Collison, CEO of Stripe, joins economist Russ Roberts to examine whether the pace of scientific and technological discovery is actually accelerating despite massive growth in the research enterprise. Collison presents a thesis grounded in multiple independent lines of evidence: while aggregate scientific progress may be holding steady or growing, the productivity of science measured per researcher, per dollar, and per hour is declining sharply and consistently across fields. He draws on work by Ben Jones and Bruce Weinberg on Nobel Prize winner ages, analysis of TFP trends over the past 150 years, and the paper "Are Ideas Getting Harder to Find?" by Nick Bloom, Chad Jones, John Van Reenen, and Michael Webb showing that productivity within specific fields like drug discovery and semiconductors is declining exponentially. The conversation examines why—not because the frontier of knowledge is exhausted, but because institutional structures create Institutional Sclerosis: conservative funding mechanisms, risk-averse committees, and what Collison calls a "Monoculture" in how science is organized.

Rather than simply documenting decline, Collison argues that much of the problem is fixable through structural reform. He points to natural experiments like the Howard Hughes Medical Institute, which funds researchers rather than projects and achieves dramatically higher citation impact than comparable NIH-funded work—evidence that changing institutional variables yields outsized returns. The conversation explores what mechanism might unlock innovation: Variance Maximization through diverse funding sources and Portfolio of Bets approaches, drawing lessons from venture capital's success in software and IT. Roberts and Collison debate whether this reflects genuine institutional failure or simply the cost of reaching harder frontiers, and what experiments in science funding and organization might reveal. The discussion ranges from the empirical case for productivity decline to the philosophical question of whether top-down institutional design can ever escape its own biases—a tension Collison frames through Seeing Like a State's critique of Monoculture.

Sharpest takeaway

Collison argues that while aggregate scientific progress may be holding steady, the productivity of science per researcher, dollar, and hour is declining sharply and at exponential rates across fields, and that a significant part of the cause is institutional monoculture rather than the depletion of low-hanging fruit—meaning we should experiment aggressively with new funding and organizational mechanisms.

  • Multiple independent lines of evidence (TFP, within-field productivity, scientist demographics, Nobel significance) all point toward declining per-person research productivity.
  • Inputs (number of scientists, hours, dollars) have grown 10x-100x without proportional gains in output.
  • Natural experiments like HHMI funding show large effect sizes from changing institutional variables, suggesting much of the problem is fixable structure, not exhausted reality.

The argument · threads4 threads · 39 claims
0.85

Institutional organization and incentive structures breed caution that stifles rather than enables discovery.

3 pointscentrality 5/5
  • The dramatic increase in published scientific articles tells us little about real discovery because there is an enormous return to publishing, so academics have innovated to increase the supply of papers in response to incentives, and similarly collaborative co-authorship may be inflated by incentives to share publication credit.

    “there's an enormous return to publishing an article. So, academics have innovated to increase the supply of that precious thing in response to those incentives.”

  • As scientific knowledge has become more organized and institutionalized, a kind of sclerosis sets in: the Nobel Prize has become so brand-conscious that awarding it to something that turns out wrong would be embarrassing, making it conservative and cautious, and the general lesson is that more organization is not always better because groupthink, caution, and risk-aversion can wring creativity out of the system.

    “there's a sclerosis, it seems to me, about scientific knowledge as it's become more organized and more institutionalized.”

  • Einstein did his innovations as a patent clerk without a large lab, but today much scientific innovation requires capital, which requires funding, which requires channeling wealth—often through the political process via NSF and NIH—and as any industry matures it becomes less eager to take risks, creating sclerosis at both the institutional and personal level.

    “one thing that's obvious is that a lot of scientific innovation today requires capital. And, once it requires capital, it requires funding. And once it requires funding, it requires wealth.”

0.75

Progress requires discovering fundamentally new approaches, not optimizing existing institutional or educational structures.

2 pointscentrality 5/5
  • Mariana Mazzucato argues the government should actively innovate—not merely fund research but actually do the innovation within government itself—an approach Roberts presents as a contrasting alternative he disagrees with.

    “She wants the government to be actively innovating--not just funding, but actually doing the innovation within government.”

  • The whole idea of finding 'the right way' to structure science funding or reform education is a dead end, because most of the time the right way is something not yet thought of; the only reliable knowledge is how not to do it—national curricula, mandatory education credentials—so progress requires genuinely new things, not doing existing things better.

    “that whole idea is just a dead end... The whole idea of finding the right way. And that most of the time, the right way is something we haven't even thought about at all.”

0.61

Per-researcher productivity falls due to picked low-hanging fruit, poorly organized knowledge search, and societal constraints on innovation.

2 pointscentrality 4/5
  • Possible explanations for declining per-person productivity fall into three classes: the nature of reality (low-hanging fruit picked, and lowest-hanging scientists already chosen), poorly organized search for knowledge (universities and labs needing overhaul), and bad rules of the game in society (legislation and cultural norms hampering productivity)—and these factors get tangled together.

    “One answer... is that this is the nature of reality. The lowest-hanging fruit's been picked... The second possibility would be, we organize the search for knowledge poorly... And the third would be we need to change the rules of the game in the larger society.”

  • Declining per-scholar productivity does not mean innovation has stopped; in the Moore's Law case, chip capacity keeps improving, but it takes an enormously larger number of researchers to produce each improvement, so it is the per-researcher rate that is falling, not the overall level of innovation in the field.

    “We have a lot more capacity on our chips, our computer chips. The problem is, is that it takes an enormously larger number of workers and researchers and scholars to produce that improvement”

0.47

Reducing public accountability for funding decisions could counteract institutional risk-aversion and unlock latent creative capacity.

2 pointscentrality 3/5
  • In Shakespeare's time there were probably a hundred potential Shakespeares who died young, went hungry, or lacked the means to write plays all day, whereas modern affluence lets thousands work in entertainment—so today's golden age of storytelling (long, high-quality television like The Americans, The Crown, The Wire) reflects opportunity unlocking latent talent.

    “there were probably a hundred Shakespeares available to be productive during Shakespeare's time, but most of them died young or spent their time hungry”

  • If cultural risk-aversion is the constraint, making discretionary funding decisions anonymous—revealing the funder's identity only if the bet works, after which they are showered with credit—could counteract the caution that comes with being held publicly accountable for risky allocations.

    “having that process be anonymous rather than named-- ... would be an interesting model.”