a16zTaking Bold Bets: NIH and the Future of Biomedical Science
CHAPTERS
- 0:00 – 0:58
Why “trust in science” requires respect, transparency, and learning from failure
Bhattacharya opens with a core thesis: the public is capable of understanding evidence when it’s presented honestly and respectfully. He argues science should adopt more of a Silicon Valley mindset—tolerating productive failure and publishing what didn’t work—so the system learns faster and trust improves.
- •Public trust follows transparent evidence, not authority
- •Respectful disagreement is healthy for scientific credibility
- •Science should stop punishing productive failure
- •Publishing negative/failed results would accelerate learning
- •“Silicon Valley spirit” as a cultural model for research
- 0:58 – 1:58
Major announcement: $50M Autism Data Science Initiative and selected teams
Bhattacharya describes being tasked with producing better answers for families affected by autism amid rising prevalence. He announces a new $50M initiative, with 13 teams selected from 250 applicants, aiming to generate higher-quality, actionable autism research over the next few years.
- •Autism prevalence has risen for decades; families lack clear answers
- •New $50M Autism Data Science Initiative launched
- •250 teams applied; 13 teams selected for large grants
- •Focus on rigorous science that can translate into prevention/treatment insights
- •Cross-agency coordination implied as part of broader agenda
- 1:58 – 4:35
Clinical updates: Leucovorin for a subset of autism and caution on Tylenol in pregnancy
He highlights two related policy/clinical items: broader access/payment for Leucovorin (folinic acid) for children with specific folate-processing issues, and emerging evidence linking acetaminophen use in pregnancy with later autism diagnoses. He stresses caution and prudence rather than panic, alongside forthcoming guidance updates.
- •Leucovorin can help some autistic children with folate-processing problems
- •Reported outcomes: speech restoration in ~20%, improvement up to ~60% (subset-specific)
- •Emerging, controversial evidence on acetaminophen in pregnancy and later autism correlation
- •Message: careful, limited use during pregnancy—avoid panic
- •FDA/CMS actions: updated guidance and payment changes
- 4:35 – 6:44
Preterm birth initiative and the broader mandate to answer families’ health concerns
Agarwala raises preterm birth as another focus area, and Bhattacharya notes U.S. outcomes lag Europe’s with incomplete explanations. He frames NIH’s responsibility as producing rigorous science that addresses real-world worries—while acknowledging how confusing and shifting scientific guidance can be.
- •U.S. preterm birth outcomes worse than Europe’s
- •Prenatal care matters but doesn’t fully explain differences
- •NIH’s role: deliver rigorous, actionable answers families are asking for
- •Science evolves (e.g., nutrition guidance like eggs) and can confuse the public
- •Need to raise standards so guidance is more reliable over time
- 6:44 – 9:39
The replication crisis: why “published” doesn’t mean “true”
Bhattacharya argues that replication should be science’s standard for truth, not journal publication or authority. He attributes replication problems to the difficulty of science, the scale/specialization of modern research, weak incentives to verify others’ work, and publication standards that are too low.
- •Replication by independent teams should be the truth standard
- •Specialization and sheer volume reduce cross-checking
- •Career incentives discourage replication work
- •Publication and peer review are insufficient signals of correctness
- •Scientists are prone to self-conviction; replication counterbalances bias
- 9:39 – 11:09
Reforms and early tenure reflections: audits, centralized review, and fixing “misread” intent
He reviews early changes as NIH director, including tightening oversight of foreign collaborations to ensure auditable spending, and centralizing grant review through the Center for Scientific Review. He notes frustration with media narratives that interpret oversight as opposition to collaboration.
- •Foreign collaborations remain important but must be auditable
- •Example: inability to audit funds sent to Wuhan lab under old system
- •New system aims to track spending and verify scientific work products
- •Centralizing review to avoid parallel institute review systems
- •Communications challenge: reforms can be misconstrued as anti-collaboration
- 11:09 – 12:47
Bringing venture-style portfolio thinking to NIH: bold bets, faster novelty, less conservatism
Bhattacharya explains how NIH has grown more conservative—funding older ideas and favoring incrementalism—reducing breakthroughs per dollar. He proposes portfolio-style tolerance for failure, plus institutional encouragement to try newer, riskier ideas and publish lessons from failures.
- •Venture analogy: most bets fail, a few win big—portfolio matters
- •NIH historically funded younger ideas (’80s/’90s) vs older ideas (2000s/2010s)
- •Conservative peer review and incentives push incremental work
- •Allow and learn from productive failure; publish what didn’t work
- •Goal: more breakthroughs per dollar and more transformative progress
- 12:47 – 14:26
How peer review becomes risk-averse and self-protective—and how to counter it
He describes structural bias in peer review: reviewers are often method experts invested in incumbent ideas and may dismiss disruptive proposals. The discussion focuses on how to create room for experimentation despite high failure rates inherent to novel science.
- •Peer review can favor incumbent paradigms and penalize challengers
- •New ideas often look implausible to specialists trained in older approaches
- •Most new ideas will fail—funding systems must expect that
- •Without room for experimentation, big advances become rare
- •Cultural and incentive changes are needed, not just procedural tweaks
- 14:26 – 23:03
Allocation vs execution: who decides NIH priorities, and why politics and science must interact
Agarwala frames NIH as needing to ‘nail’ resource allocation across disease areas and execute well within each. Bhattacharya argues allocation inevitably reflects democratic will (Congress/president) alongside scientific opportunity, while scientists should guide within-area portfolio choices; he cites HIV activism as a case where political pressure corrected under-response.
- •Two-level problem: macro allocation across diseases vs micro execution within fields
- •Congress/president decide budgets; NIH mediates scientific opportunity and public need
- •HIV history: political mobilization helped drive needed research attention
- •No “philosopher king” can perfectly allocate; public voice matters
- •NIH should inform Congress about emerging high-leverage opportunities
- 23:03 – 26:33
America’s stalled life expectancy and the shift toward chronic disease priorities
Asked about under/over-allocation, he argues the key is not just more money but better alignment with major health burdens. He highlights chronic disease, rising cancer incidence, diabetes, kidney failure, and autism as areas where better translation from science to health outcomes is needed, connecting this to broader public movements demanding change.
- •U.S. life expectancy has stagnated for over a decade
- •Chronic diseases and rising burdens: diabetes, cardiovascular disease, kidney failure, autism
- •Cancer: better survival but rising incidence
- •NIH should focus on practical health needs where suffering is greatest
- •Public pressure/‘movement’ creates opportunity to reform priorities and delivery
- 26:33 – 35:46
Fixing the early-career investigator bottleneck: age, incentives, mentorship, and institute accountability
The conversation turns to the aging of first-time major grant recipients and the loss of younger innovators. Bhattacharya outlines plans to judge institutes on portfolio outcomes, align funding with strategic plans, increase support for early-career investigators, and reward senior investigators who effectively mentor and advance junior scientists.
- •Median age for first major NIH grant rose from ~35 to mid-40s
- •Extended postdoc culture delays independence and discourages talent retention
- •Younger scientists disproportionately generate newer ideas
- •Institute directors should be evaluated on portfolio impact, not single-grant success
- •Reward mentorship and career advancement of early-career scientists within grants
- 35:46 – 38:29
Academic freedom and open science: removing internal permission barriers and challenging publishing gatekeepers
Bhattacharya describes changing rules so NIH intramural researchers no longer need supervisor permission to publish. He criticizes concentrated, for-profit journal control and paywalls on publicly funded research, noting NIH has removed paywalls for NIH-funded work and wants further openness and freedom in publishing.
- •NIH intramural researchers no longer need supervisor permission to publish
- •Academic freedom is framed as essential to excellent science
- •Concern about a publishing ‘duopoly’ and high fees charged for public-funded research
- •Paywalls for NIH-funded papers removed for public access
- •Further policy work planned to expand openness and publishing reform
- 38:29 – 41:52
Rebuilding trust after the pandemic: gold-standard science, humility, and a servant posture to the public
He attributes the trust collapse to pandemic-era policies perceived as unscientific or inconsistent (e.g., plexiglass, masking rituals, school closures). His proposed path forward combines rigorous standards (replication, unbiased review, humility) with a partnership model where public health serves rather than dictates to the public.
- •Pandemic policies undermined credibility: plexiglass, inconsistent rules, school closures
- •Trust rebuild requires ‘gold-standard’ science and reproducibility
- •Humility about limits and uncertainty is essential
- •Public health should act as a servant/partner, not an authority above people
- •Trust restoration will take time due to deep, widespread skepticism
- 41:52 – 48:58
Communicating uncertainty without blaming the public—and when confidence is justified
Bhattacharya argues officials should plainly say “I don’t know” when evidence is weak, analogous to a medical trainee resisting the urge to bluff. He distinguishes uncertainty from higher-confidence areas (e.g., MMR), emphasizing open debate, evidence visibility, and the belief that the public responds rationally when treated respectfully.
- •Honest ‘I don’t know’ prevents overconfident policy mistakes
- •Pandemic-era certainty on weak evidence damaged lives and credibility
- •Avoid framing distrust as public ignorance; scientists must communicate better
- •Some recommendations warrant high confidence (e.g., MMR’s benefit for measles prevention)
- •Open debate + replicated evidence can win trust over time
- 48:58 – 51:51
NIH priorities: chronic disease breakthroughs and integrating AI—promise plus guardrails
He highlights promising chronic disease angles (e.g., shingles vaccine correlations with reduced Alzheimer’s risk) and argues NIH should give ‘scientific love’ to scalable, high-impact ideas. On AI, he cites AlphaFold, clinical workflow improvements, and diagnostic support, while warning about hallucinations and the need for research and safe deployment.
- •Example opportunity: observational evidence shingles vaccine may reduce Alzheimer’s decline risk
- •Portfolio should prioritize scalable, practical chronic disease wins
- •AI accelerates drug discovery (e.g., protein folding) and focuses wet-lab efforts
- •Clinical AI use cases: radiology support, ambient scribes for EHR burden
- •Need research, privacy protections, and safeguards against AI errors/hallucinations
- 51:51 – 54:14
Limits of AI in science operations: noise, grant floods, and AI as augmentation—not replacement
Agarwala asks whether AI could write roadmaps, submit grants, and review proposals; Bhattacharya rejects a fully automated future. He describes curbing excessive, AI-generated grant submissions that overwhelm review capacity, and positions internal secure AI tools as productivity multipliers rather than substitutes for scientific creativity.
- •AI is strong at summarizing existing knowledge, weaker at paradigm-breaking novelty
- •Problem: high-volume AI-generated grant submissions create noise
- •Policy response: limits on number of new applications per cycle
- •Secure enterprise AI tools can help with NIH/HHS tasks under privacy constraints
- •Overall framing: AI augments scientists; it doesn’t replace them
- 54:14 – 58:59
Advice to scientists and the ‘Max Perutz’ story: persistence, big ideas, and an NIH that enables them
Bhattacharya closes by encouraging scientists to stay in research and persist despite rejection. He tells the story of Max Perutz’s long, discouraged path to solving protein structure—eventually transformative—and frames NIH’s mission as building an ecosystem where modern Perutz-like innovators can take real swings.
- •Science can dramatically advance human well-being—worth the struggle
- •Progress often comes from persistent individuals despite repeated rejection
- •Max Perutz as exemplar: decade-long push against skepticism led to breakthrough
- •Question posed: would today’s system allow such long-horizon, high-risk work?
- •NIH should enable bold experimentation that can change the world
