Episode Summary
Executive Summary: Rob Wiblin interviews Spencer Greenberg about his work building scalable tools for mental health, behavior change, and better social science. The conversation ranges from online research platforms and CBT-based apps to scientific incentives, entrepreneurship, AI risk, and surveys revealing public misunderstandings about charities and animal welfare.
Main Topics: Scalable mental health and behavior-change products: Greenberg describes SparkWave’s apps, especially Uplift for depression and planned anxiety/sleep tools, built around evidence-based CBT and iterative testing. Improving social science methodology: He argues online recruitment, better study design, withholding data for out-of-sample tests, and iterative research can reduce false positives and improve reliability. Clearer Thinking and decision-making tools: Greenberg explains how clearerthinking.org turns behavioral science into interactive tests and training, including rationality and misconceptions assessments. Entrepreneurship and startup impact: He discusses traits linked to startup success, especially persistence plus flexibility, and how startups can unintentionally do harm through misinformation, irrationality, or externalities. AI, catastrophic risk, and public perception: The interview covers automation, power concentration, and misalignment risks from advanced AI, plus survey evidence that many people underestimate extinction risk. Survey research on public beliefs: Greenberg and Wiblin discuss surveys on charity cost-effectiveness and animal welfare attitudes, showing people often hold surprisingly inaccurate or conflicted views. Effective altruism and cause prioritization: Greenberg praises effective altruism but urges broader cause exploration, including more direct scientific research and neglected areas like biology.
Key Arguments: Online research tools can make studies vastly cheaper, faster, and more iterative than traditional lab-based social science, enabling rapid testing of behavior-change interventions. Research quality should be improved with techniques common in machine learning and finance, especially out-of-sample validation and holding back data before final hypothesis checking. Academic incentives favor publishable novelty over truth-seeking, which can encourage false positives, underpowered studies, and low replicability. Mental health interventions should be designed to scale widely; software can deliver CBT-like support cheaply and potentially globally. Startups can create harm even when users buy products, via misinformation, exploitation of irrationality, zero-sum dynamics, or negative externalities. A good founder needs both persistence and flexibility: not giving up easily, but being willing to change strategy when evidence demands it. AI poses three major classes of risk: labor displacement, concentration of economic and strategic power, and loss of human control over highly capable systems. Many people and even effective altruists severely underestimate catastrophic risks such as human extinction, suggesting a need for more attention to long-tail global threats. Public attitudes toward charity and animal welfare are often poorly calibrated; people underestimate how cost-effective top charities can be and show mixed reasoning about meat and farm animal suffering.
Data Points: Uplift early outcomes: 50% reduction in depression among the first 80 completers - Greenberg describes results from SparkWave’s depression app study over about a month Uplift follow-up durability: Benefits largely maintained at 6 weeks and 6 months - Follow-up assessments suggested the initial improvements persisted Mental health app format: 1-2 sessions per week, 30-40 minutes each - Describing how the CBT-based Uplift program works Habit interventions tested: 22 micro-habit formation interventions - SparkWave randomized combinations of small habit-change techniques Clearer Thinking mailing list: About 20,000 people - Audience size for clearerthinking.org Clearer Thinking usage: Over 150,000 people - Users of the free decision-making tools and training programs Task Recruiter group profile: Skews younger, around 30s to 50s - Why online samples can still fit target audiences for certain products Public extinction risk estimate: Near-zero / essentially impossible - General public survey responses about human extinction in the next 50 years EA extinction risk estimate: About 100,000 times higher than the general public’s estimate - Survey comparison between effective altruists and general population AI causing 10% population loss in 50 years: EA median 10%; general public 1% - Survey of catastrophic risk perceptions Major nuclear war in next 50 years: Around 10% across groups - Survey result discussed as worryingly high Charity cost to save a life: Median $40 for a typical charity; $25 for the most cost-effective charity - Public survey of beliefs about aid effectiveness Perceived difference in charity effectiveness: People thought most effective charities were only slightly better than typical ones - Shows underestimation of variation in cost-effectiveness Animal pain/suffering belief: 99% said animals can experience pain; 99% said animals can suffer - Survey result on animal sentience beliefs Moral wrongness of hurting animals for enjoyment: 88% said it is wrong - General question about hurting animals to enjoy life more Moral wrongness of hurting farm animals: 65% said it is wrong; 21% unsure; 13% said no - Farm-animal-specific survey question Hurting animals for taste: 53% said it is wrong; 20% unsure - Question about hurting animals mainly because they taste good Startup team/product count: About 11 products - SparkWave’s portfolio of projects Machine learning work: Used in some SparkWave products - Greenberg notes several products use ML, though not named publicly
Pivotal Quotes: "The fundamental issue that's the core of everything that you said, that all these different problems, they really all emanate from the same source, which is that the incentive alignment is just not there." — Spencer Greenberg: On why academic social science often produces weak or misleading findings "The key is not under the lamp post, then you're just in a very tricky situation, whatever you do. Exactly. And that's how I think about research." — Spencer Greenberg: Explaining why researchers should first pursue questions that are easier and cheaper to study "The reason you want a lot of variables is because it gives you a lot of flexibility to ask questions you hadn't even thought of, to check things in different ways... but you want to hold yourself to a really high standard by doing things like withholding data in a vault." — Spencer Greenberg: On combining exploratory research with rigorous validation
Implications: Listeners are encouraged to think more empirically about careers, products, and causes: build scalable tools, test them rigorously, distrust weak incentives, and pay more attention to underappreciated risks like AI and animal suffering.