Patrick Collison on Innovation and Scientific Progress

Episode Summary

Executive Summary: Russ Roberts and Patrick Collison discuss Collison’s Atlantic essay arguing that scientific progress appears to be slowing on a per-person, per-dollar basis even as total research effort rises. They review evidence from productivity trends, field-level outputs, Nobel Prize analyses, and institutional incentives, then explore explanations ranging from hard-to-find “low-hanging fruit” to funding-system sclerosis, risk aversion, and lost opportunity for unconventional talent.

Main Topics: Stripe as internet infrastructure (Priority: 4/5): Collison explains Stripe’s origins as a tool to make online payments and broader financial infrastructure easier for internet businesses, emphasizing the company’s mission to make transacting universal. Is scientific progress slowing? (Priority: 5/5): The core debate centers on whether science’s overall pace is slowing or whether productivity is declining per researcher, per hour, and per dollar despite more total input. Evidence for declining research productivity (Priority: 5/5): They review macroeconomic productivity trends, field-specific studies, Nobel Prize age and team-size data, and Collison’s subjective ranking of Nobel-winning discoveries as evidence that discoveries are getting harder to generate. Institutional and cultural causes (Priority: 5/5): The discussion explores whether the slowdown reflects the nature of reality, exhausted low-hanging fruit, or problems in the scientific ecosystem such as funding conservatism, groupthink, and rigid career structures. How funding mechanisms shape innovation (Priority: 4/5): Roberts and Collison compare centralized grant systems with more decentralized, risk-tolerant models like HHMI, Y Combinator, and direct allocation by scientists to argue for experimentation in science funding. Opportunity, talent, and variance (Priority: 4/5): They connect innovation to the uneven distribution of opportunity, arguing that breakthrough work depends on environments that allow high-variance, unconventional people to flourish.

Key Arguments: Progress should be judged per researcher/per hour/per dollar, not only in aggregate, because total effort has risen enormously while individual productivity appears to have fallen. Macro evidence such as U.S. total factor productivity suggests a mid-20th-century surge followed by long-run stagnation, consistent with slower knowledge generation. Field-specific studies, especially 'Are Ideas Getting Harder to Find?', show exponentially declining productivity within domains like semiconductors, crops, and drug discovery. The rising age of Nobel winners and larger scientific teams suggest that frontier discovery requires more time, more collaboration, and more accumulated knowledge than in the past. Subjective ranking of Nobel Prize breakthroughs across eras found physics looking especially front-loaded in the 20th century, though chemistry and medicine appeared flatter. Centralized science funding may be too conservative, too standardized, and too slow to support unconventional researchers or high-risk ideas. Scientific institutions may reduce variance rather than increase it, which is a problem if breakthroughs are 'positive catastrophes' that come from rare outliers. More decentralized, judgment-based funding can have large effects; HHMI-style support and early-career independence may outperform rigid grant systems. Talent is widely distributed, but opportunity is not; expanding access can unlock breakthroughs that current institutions miss. The right response is not simply to spend more on science, but to reallocate funding and experiment with alternative institutional designs.

Data Points: Date of episode: December 14, 2018 - EconTalk introduction to the interview with Patrick Collison. U.S. GDP per capita vs self-reported well-being: R squared = 0.68 - Collison cites this as evidence that growth correlates strongly with well-being. Average Nobel laureate age (early days): 37 years old - Ben Jones and Bruce Weinberg finding on Nobel Prize winners in earlier eras. Average Nobel laureate age (today): 47 years old - Same analysis showing winners are about 10 years older now. Increase in Nobel age: +10 years - Used to suggest it takes longer to reach frontier knowledge. Researcher growth: 10x to 100x more scientists - Collison argues the increase in working scientists is enormous relative to the difficulty of detecting proportional output gains. HHMI funding effect: 98% more likely - Azoulay study finding HHMI recipients were 98% more likely to produce top 1% cited work. Top-cited work threshold: Top 1% by citations - Outcome used in the HHMI funding study. Team size / authorship trend: Substantially growing - Collison notes larger scientific teams and more coauthorship as part of the evidence base. Aggregate research growth: Roughly constant or still rising - Collison’s claim that total progress may be stable even if per-person productivity is falling.

Pivotal Quotes: "what we are much more confident of ... is that the rate of progress is declining very rapidly on a per person, per hour, or on a per dollar basis" — Patrick Collison: Defines the central thesis of the Atlantic essay and the episode. "there are all these mechanisms that actually seem not only not designed to increase the variance, but in many ways that they kind of structurally attenuate it" — Patrick Collison: On why scientific institutions may suppress the rare outlier discoveries needed for major breakthroughs. "we do know that the number of working scientists has increased enormously" — Patrick Collison: Used to argue that if inputs rose 10x-100x, output should have risen comparably if productivity were unchanged.

Implications: If scientific productivity is slowing per researcher, institutions should focus less on simply adding scientists and more on reworking incentives, funding, and career paths to support high-variance talent and unconventional ideas.