Open-ended research delivers novel gains in the AI era
AI can speed up experiments, search papers, and predict candidates. But it can't ask the weird, unpopular questions that lead to step-change discoveries. That still comes from curiosity with no guarantee of payoff.
Japan's push for immediate returns risks starving this curiosity. The recent chemistry Nobel for Susumu Kitagawa, whose work on porous framework materials took decades to mature, is a reminder: big wins often start as open-ended bets, not quarterly projects. See the pattern across biology and materials science-breakthroughs often trace back to years of patient, basic inquiry.
What short-termism costs
- Safer topics, smaller ideas: funding chases near-term publications and patents, not high-variance questions.
- Lab behavior skews to KPI gaming: short grants and crowded panels reward consensus and familiar methods.
- Translation gets thinner: without deep wells of basic insight, there's less to commercialize later.
- Talent opts out: the best minds don't want to spend careers formatting grant PDFs for incremental work.
A smarter portfolio for institutions
- Ring-fence 15-20% for open-ended projects with 5-10 year horizons. Protect this budget from annual cuts.
- Fund people, not just projects: portable, multi-year fellowships that buy time and independence.
- Adopt "safe-to-fail" experiments: small stakes, high-variance trials with fast kills and fast rebets.
- Evaluate by learning, not just outputs: pre-register bold hypotheses, track what was learned even if the result was null.
- Spin-in, not just spin-out: let applied teams pose hard questions back to basic labs and co-fund the search.
Policy moves Japan can make now
- Create 10-year, renew-by-milestone grants for curiosity-driven work. Review panels should include builders who respect technical risk.
- Set a national open-ended research quota across agencies and universities. Publish it, then raise it.
- Simplify admin: one-page renewals, lighter reporting, fewer compliance traps that divert researchers from the bench.
- Offer compute and lab credits as funding. Don't force basic research to choose between chemicals and servers.
- Reward cross-discipline bets: chemistry x AI, immunology x materials. Give joint PIs shared credit and budgets.
- Update metrics: prize long-horizon citations, data sets, negative results, and tools-signals that compound over time.
AI is an amplifier, not a replacement for curiosity
AI helps researchers scan literature, generate candidates, and sift data. It makes exploration cheaper. But it still needs human taste for questions that look odd at first and obvious later.
The lesson from Japan's standout successes in chemistry and immunology is clear: bet on independent minds early, then give them time and room. AI will make those bets pay faster, but the bets must exist in the first place.
Practical steps for lab heads this quarter
- Reserve one meeting a month for "wild ideas only." No slides. Just the question, the path, and the kill criteria.
- Write a one-page personal research memo: your 3 biggest questions for the next 5 years and why they matter.
- Split your group time 70/20/10: core deliverables, adjacent probes, speculative shots. Defend the 10.
- Pre-commit to two external collaborations that force you into unfamiliar methods.
For funders and boards
- Balance the R&D book: at least 15% long-horizon basic research, 35% platform building, 50% near-term delivery.
- Back people with taste. Track their questions and pivots, not just the h-index.
- Tie a slice of executive incentives to long-term research health: retention of early-career PIs, reproducible data assets, and shared infrastructure built.
Breakthroughs look wasteful until they work, then inevitable after. If Japan keeps demanding immediate proof, it will keep getting immediate results-and miss the ones that change the arc of entire fields.
Background on long-term science funding models and recognition can be found at Nobel Prize in Chemistry and the OECD's guidance on research measurement in the Frascati Manual.
If you're updating lab workflows with AI to free time for deeper questions, see curated programs at Complete AI Training.
Your membership also unlocks:
