Research Funding & Pace
Each new idea takes more researchers and money, while public funding swings politically: how much to fund, and how, stays open.
Open in the interactive tree →Bloom, Jones, Van Reenen and Webb showed (2020) that ever more researchers are needed to keep productivity growth constant. Park, Leahey and Funk (Nature, 2023) found papers and patents becoming less 'disruptive' over decades. Applying for grants is costly: an Australian study estimated 550 working years of researcher time for one round of 3,727 proposals, of which 21% were funded.
As of October 2026
In the US, the administration proposed cutting NIH by about 40% for fiscal 2026 and froze or terminated many grants; by January 2026 about 2,600 grants worth roughly $1.4 billion were still suspended (per Wikipedia). Congress rejected the cut and set NIH funding at $48.7 billion (up $415 million), and in January 2026 an appeals court upheld the block on the 15% cap on indirect costs. For fiscal 2027 the President again requested a cut of more than 10%, the House committee rejected it in June, and in September Congress passed a stopgap keeping funding at 2026 levels until 11 December. Funding stability, not only funding level, has become a planning risk for research.
What is missing
- Multi-year, plannable funding that does not depend on election cycles
- Leaner grant procedures and more room for high-risk ideas (e.g. lotteries, ARIA/DARPA-style models)
- Meaningful measures of scientific output instead of citation counts
- Use of AI and automation to cut researcher time per idea
- International sharing of costly large facilities
Becomes possible once solved
- Faster discovery cycles
- More research per euro spent
- Stable career paths for young researchers
Open steps
- Researcher time per idea Medium AI leverageCut the researcher time needed to test one idea with AI that runs literature, analysis and drafting cycles.
- Cheaper, fairer grant selection Low AI leverageCompare panel review, lotteries and partial randomization on cost, bias and later research output.
- Measuring scientific output Medium AI leverageMeaningful measures of scientific output instead of citation counts, tested against later impact.
- Cost of stop-and-go funding Low AI leverageEstimate how funding swings and grant freezes change research output, to show what stable funding is worth.
Where AI could help
Low AI leverage. Budgets and funding stability are set by governments and funders; AI can cut researcher time per idea and review effort but cannot set money.
- AI scientist agents that run literature and data-analysis cycles and cut researcher time per hypothesis
- Screening grant proposals for eligibility and matching reviewers
- Automating routine analysis, coding and writing so grant effort per result drops
- Mapping the literature to spot under-funded, high-impact gaps
Shown so far
- In February 2025 Google described an AI co-scientist whose hypotheses on leukemia drug repurposing and liver fibrosis targets were tested in lab experiments (Google-authored report). source
- In November 2025 Edison Scientific reported that its Kosmos AI scientist reads about 1,500 papers per run and that independent scientists rated 79.4% of report statements accurate (company-reported). source
Prerequisites
- Peer Review & Research Bodies1945-1975
- AI as Research Partner2025-2026