Custodes Futurifor the advancement of humanity

Economic factor, barrier 6 of 6

Slowing research productivity

Research productivity is declining. Keeping computer chip progress on pace now requires far more researchers than in the early 1970s (Bloom et al., 2020).

Evidence

  • Chip density. Researchers needed to keep doubling chip density rose to more than 18 times the number needed in the early 1970s (Bloom et al., 2020).
  • Burden of knowledge. As knowledge accumulates, age at first invention, specialization, and teamwork all rise (Jones, 2009).
  • Critique. Critics question the measures of idea output used in the productivity studies, and the authors have responded (Bloom et al., 2020).

Proposed and experimental methods

Methods that are proposed, under trial, approved in some places, or tried and then failed. Each shows a stage label and an evidence rating. A stage label shows how far a method has progressed, not whether it works. The stage labels are explained on the economic factor page.

  • Funding people for the long term instead of short projects (Approved but not scaled, B). Howard Hughes Medical Institute investigators, who receive long-term support that tolerates early failure, produced high-impact papers "at a much higher rate" than similarly accomplished NIH-funded scientists with short review cycles, and shifted toward more novel lines of research (Azoulay, Graff Zivin, and Manso, 2011). The comparison relies on statistical matching, not random assignment, and HHMI selects very strong scientists.
  • ARPA-style agencies (Approved but not scaled, B). These give program managers wide freedom to design programs, pick projects, and stop those that miss milestones (Azoulay et al., 2019). Among 2010 cleantech startups, the 25 funded by the U.S. energy agency ARPA-E patented more than all comparison groups and did better in business than rejected applicants, but not better than other cleantech startups (Goldstein et al., 2020), and reviewers noted ARPA-E had yet to show DARPA-scale wins (Azoulay et al., 2019). Newer copies include the UK's ARIA, created by an Act passed on February 24, 2022 (UK Parliament, 2022), and the U.S. health agency ARPA-H, given $1 billion on March 15, 2022 for its first three years (ARPA-H, n.d.). Neither has published outcome evaluations.
  • Partial lottery grant funding (Approved but not scaled, C). Proposals that pass a quality screen are funded at random, which aims to cut reviewer bias and wasted effort. New Zealand's Health Research Council funds its Explorer Grants (NZD 150,000 for up to two years, about 2% of its spending) this way, and in a survey 63% of applicants found randomization acceptable for these grants and 25% did not, while only 40% favored it for other grants, with a 39% response rate (Liu et al., 2020). No study yet compares the research produced with that from peer review alone.
  • Very fast grants (Early trial, C). During the COVID-19 pandemic, Fast Grants raised over $50 million and made over 260 grants from more than 6,000 applications, with first-round decisions within 48 hours. In the founders' survey of grantees, 64% said the work would not have happened without the grant and 78% said they would change their research program "a lot" if funding were unconstrained (Collison, Cowen, and Hsu, 2021). The evidence is self-reported and comes from the program's own founders.
  • Focused research organizations (Theoretical, C). Proposed nonprofit start-ups that take on mid-scale, time-limited projects, such as building a tool or dataset, that academia, venture capital, and government labs leave undone (Marblestone et al., 2022). No outcome evaluation was found.
  • AI tools for science (In wide use, B). The AlphaFold system predicted protein structures with accuracy competitive with experimental structures in a majority of cases in the CASP14 assessment, a task that had been an open problem for more than 50 years (Jumper et al., 2021). Whether such tools raise overall research productivity has not been measured well (see the next entry).
  • Not shown to help: Claimed evidence that AI raised discovery in corporate research (Failed or reversed, D). A widely discussed 2024 preprint, "Artificial Intelligence, Scientific Discovery, and Product Innovation," reported large gains from an AI tool in a research lab. In May 2025, MIT stated after an internal review that it had no confidence in the provenance, reliability, or validity of the data and asked for the paper to be withdrawn from arXiv and from the journal to which it had been submitted (MIT Economics, 2025).

Sources cited on this page

  1. Bloom, N., Jones, C. I., Van Reenen, J., & Webb, M. (2020). Are ideas getting harder to find? American Economic Review, 110(4), 1104-1144. AEA B Moderate
  2. Azoulay, P., Fuchs, E., Goldstein, A., & Kearney, M. (2019). Funding breakthrough research: Promises and challenges of the "ARPA model". In J. Lerner & S. Stern (Eds.), Innovation Policy and the Economy, 19. University of Chicago Press. link B Moderate
  3. Azoulay, P., Graff Zivin, J. S., & Manso, G. (2011). Incentives and creativity: Evidence from the academic life sciences. RAND Journal of Economics, 42(3), 527-554. link B Moderate
  4. Collison, P., Cowen, T., & Hsu, P. (2021, June 15). What we learned doing Fast Grants. Future. link C Limited
  5. Goldstein, A., Doblinger, C., Baker, E., & Anadón, L. D. (2020). Patenting and business outcomes for cleantech startups funded by the Advanced Research Projects Agency-Energy. Nature Energy, 5, 803-810. link B Moderate
  6. Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature, 596(7873), 583-589. link B Moderate: (A for structure prediction accuracy)
  7. Liu, M., Choy, V., Clarke, P., Barnett, A., Blakely, T., & Pomeroy, L. (2020). The acceptability of using a lottery to allocate research funding: A survey of applicants. Research Integrity and Peer Review, 5, 3. link C Limited
  8. Marblestone, A., Gamick, A., Kalil, T., Martin, C., Cvitkovic, M., & Rodriques, S. G. (2022). Unblock research bottlenecks with non-profit start-ups. Nature, 601(7892), 188-190. link C Limited

Every source for this factor is listed on the economic factor page.