Interview
#41 - David Roodman on incarceration, geomagnetic storms, & becoming a world-class researcher
- Plans to engage in cause prioritization discussions weighing animal versus human welfare and present versus far future, with an anticipated near-term shift to reviewing AI safety as a primary focus for minimizing harm.
- Expects Open Philanthropy to transition from pilot mode to making firm allocation decisions regarding catastrophic risks versus direct giving if annual funding scales to $200 million or more.
- Anticipates that scaling significant funding will necessitate greater scrutiny of published research before reliance, citing current peer review as suboptimal and predicting future trends toward mandatory public data sharing, code transparency, and pre-registration.
- Estimates the probability of a Carrington-level geomagnetic storm recurring within the next decade at zero to four percent, based on historical data since 1950.
- Projects that engineering research for geomagnetic storm field experiments using full-size transformers and small cities could cost tens of millions of dollars.
- Believes the 1859 Carrington event was likely at most twice the magnitude of storms occurring since 1950, using the generalized Pareto family distribution for extrapolation rather than power law assumptions.
- Cautioning that while geomagnetic storms are not a pure existential risk, they present real tail risk regarding damage to transformers that may bypass automatic grid safety mechanisms.
- Notes that young effective altruism participants possess high potential but lack experience, creating a dynamic where the movement may have intelligence without wisdom.
- Expresses concern that the "effective altruism" label generates resistance due to its perceived arrogance, and highlights tribal divides between health and economics that hinder consensus on interventions like deworming.
- Predicts that the deworming intervention is likely positive on average but possesses weak evidence, leading to a 20-30% probability of negative impact that requires Bayesian reasoning to assess.
- Identifies selection bias as the most probable reason if the deworming intervention proves ineffective, citing the Worms study as the critical evidence despite robustness issues found in the Blakely study which showed long-term convergence rather than sharp gains.
- States that future deworming research must prioritize long-term impacts over ten years, as short-term health metrics account for only a fraction of the cost-effectiveness.
- Suggests that modern machine learning and Bayesian methods are becoming essential in econometrics and policy analysis to handle big data, avoid overfitting, and manage weak evidence by fusing priors with data.
- Warns that academic and think tank environments feature high political risks where young researchers may avoid replicating studies or challenging senior work due to tenure pressures and tribal dynamics.
- Observes that think tank staff turnover is high at junior levels due to flat organizational structures, and that advancing to top research positions generally requires a PhD, though co-authorship is possible without one.
- Expects that policy analysis roles in government or think tanks offer broader exposure to interdisciplinary dimensions like equity and administration compared to academia.
- Highlights that incarceration reduces short-term crime but often increases long-term recidivism, with cost-benefit analyses and "devil's advocate" readings yielding a break-even point of roughly $92,000 per person-year of lost liberty.
- Finds that deterrence effects of tougher sentences are essentially zero at current policy margins and that the "parole bias" in recidivism studies explains why early release may appear to increase crime.
- Advocates for self-sustaining microfinance institutions focused on savings and insurance to avoid the risks of loan-created bubbles, while noting that feedback loops in tax policy can take 10 years to manifest.
- Notes that slow-mode thinking skills derived from a math background are transferable to diverse fields, and that blogging for book creation can generate immediate engagement despite slowing the writing process.
- Concludes that using utilitarianism as a common currency may help overcome tribal moral divides to solve real-world problems, even if it does not constitute a complete morality for every individual.