Episode Summary
Executive Summary: Matt Clancy argues that speeding up science is still likely net positive, even after accounting for dual-use risks like engineered pandemics. He models science as highly valuable for health and income, but potentially costly if it accelerates a “time of biological perils.” He ultimately favors “safe acceleration”: continue improving science while investing in biosafety, better institutions, and risk mitigation. The episode also covers his skepticism that AGI will cause explosive economic growth.
Main Topics: Why speed up science at all?: Clancy explains the core rationale for meta-science: even a modest boost in scientific efficiency could create enormous social value because science generates widely shared, cumulative benefits across many years. Dual-use science and the time of biological perils: The interview explores the idea that faster science can advance dangerous technologies—especially synthetic biology—by bringing forward the era in which bad actors can cause catastrophic harm. Modeling the value of science: Clancy describes a framework that estimates science’s value through income gains, health gains, and the delayed diffusion of discoveries, while discounting distant future benefits due to uncertainty. Forecasting catastrophic bio risk: A major part of the report relies on the Forecasting Research Institute’s Existential Risk Persuasion Tournament, comparing superforecasters and domain experts on genetically engineered pandemic risks. Discounting the future and extinction risk: The conversation distinguishes ordinary long-run uncertainty from extinction risk, which may justify treating some distant-future outcomes differently because extinction is effectively permanent. Policy response: safe acceleration: Rather than slowing science broadly, Clancy argues for targeted improvements to scientific institutions plus separate efforts to reduce downside risks in biosecurity and governance. Skepticism about explosive AGI-driven growth: In the second half, Clancy explains why he is more skeptical than some peers that AGI will create explosive 10%-plus annual growth, citing frictions, bottlenecks, and historical precedent.
Key Arguments: Science has historically generated very high social returns because its discoveries are non-rival, cumulative, and diffuse across generations. Accelerating science can also accelerate dangerous capabilities, especially in biology, where one actor with the right tools could cause catastrophic harm. A one-year delay in science is not trivial because the benefits of discovery often materialize decades later in technologies, medicine, and productivity. The value of science can be modeled by combining income gains, health gains, and the cost of bringing forward a risky era of biological capability. The Forecasting Research Institute tournament provides the best available quantitative input for estimating how likely engineered pandemics and extinction risks might be. Even if science does raise catastrophic bio-risk, the model still finds large net benefits unless one assigns very high probabilities to severe outcomes. Extinction risk changes the calculus because if humanity goes extinct, distant future benefits no longer matter in the same discounted way. A practical policy stance is “safe acceleration”: improve science, but pair it with targeted biosafety, governance, and institutional reforms. Clancy is skeptical that AGI will automatically produce explosive growth because real economies contain many frictions, adoption delays, and task-specific bottlenecks. Historical episodes of fast growth and technological diffusion do not necessarily support sustained 20% annual growth once frictions are accounted for.
Data Points: Annual science spending: $360 billion - Used as the baseline spending level on science in the U.S. for the ROI thought experiment. Hypothetical efficiency gain from meta-science: 10% more effective science - Illustrative assumption for how much science could improve if a billion dollars/year were invested in better institutions. Implied value from a 10% science boost: $36 billion - If $360B/year in science spending becomes 10% more effective, that yields the equivalent of $36B in extra value. Social value per dollar of science: $7 in social value per $1 - Open Philanthropy-style benchmark used to convert science gains into social welfare value. Illustrative annual value from a $1B meta-science investment: $2.5 trillion - Derived by combining a 10% improvement in science with the estimated value generated by science spending. Estimated return on science (no bio-risk): $69-$70 per $1 - Headline model estimate for the average value of one dollar spent on science, excluding major catastrophic risk costs. Return on science using superforecaster bio-risk estimates: About $68 per $1 - Including bio-risk barely changes the net return when using the more optimistic forecasting group. Return on science using domain expert bio-risk estimates: About $50 per $1 - Using the more pessimistic biosecurity experts reduces the estimated value by roughly a quarter. Break-even annual bio-risk threshold: 0.13% per year - Annual probability of death from new biological capabilities needed to make pausing science for a year look worthwhile in the model. COVID-19 excess deaths in 2021: 0.1% of the world - Used as an intuitive anchor for comparing the magnitude of modeled biological catastrophe risk. Probability difference on engineered-pandemic risk: 18x higher - Domain experts’ annual risk estimates were about 18 times higher than superforecasters’. Extinction risk estimate difference: 140x higher - The two forecasting groups diverged far more on the chance that engineered pandemics could cause human extinction. Economic growth attribution to technological progress: ~1% per year - Clancy cites estimates that about half of 2% annual U.S. GDP per capita growth comes from technological progress. Share of corporate R&D depending on public research: 20% - Survey of R&D managers used as one estimate of how much private innovation depends on publicly funded science. U.S. patents citing academic papers: 26% - Patents in 2018 that cited academic literature, used as a proxy for science’s influence on innovation. Indirect patent-science linkage: ~60% - If indirect citation chains are included, a much larger share of patents appears connected to science. Health impact assumption: 50% of health gains from science - Model assumption used to estimate how much of life-expectancy improvement is due to science rather than income or other factors. Discount rate: 2% per year - Used as an epistemic discount rate reflecting the chance the world changes so much that historical trends stop being predictive. Annual probability of regime change: 1 in 50 - Another way Clancy characterizes the 2% discount rate: about once every 50 years, the world may change enough to invalidate extrapolation. Forecasting tournament size: 169 participants - Participants in the Existential Risk Persuasion Tournament, split between superforecasters and domain experts. Bio-risk expert panel size: 14 experts - The subset of domain experts focused on biosecurity in the forecasting tournament. Number of questions in tournament: 59 questions - The forecasting exercise covered a broad set of existential-risk questions. Timeline for genetically engineered pandemic concern: After 2030 vs. before 2030 - Forecasts compared risk before and after 2030 to infer whether risk increases with time and capability. Annual probability in superforecaster model: 0.0021% - The model’s implied annual probability of death from new biological capabilities using superforecaster estimates. Annual probability in domain-expert model: 0.0385% - The model’s implied annual probability of death from new biological capabilities using domain expert estimates. Chance Clancy assigns to superforecasters being right: ~75% - He says he leans toward superforecasters over domain experts, though not with high confidence. AGI explosive growth skepticism: 10-20% chance - Clancy’s rough estimate for the probability of explosive growth from AGI.
Pivotal Quotes: "“we should keep doing it. But also, if these domain experts are right, the risks are very important and quantitatively matter a lot.”" — Matt Clancy: Summarizing his bottom line after modeling both benefits and biological risks of faster science. "“safe acceleration”" — Matt Clancy: The policy stance he and Open Philanthropy settled on: accelerate science while also mitigating downside risks. "“I think that if you look historically, 2% seems like we’re like something changing every 50 years.”" — Matt Clancy: Explaining the epistemic discount rate and why historical trends may stop being useful over long horizons.
Implications: The episode argues for improving science, but with tighter biosafety, better institutions, and risk-aware grantmaking. For AI, it suggests caution about assuming automatic explosive growth; frictions may keep progress substantial but not runaway.