Episode Summary
Executive Summary: Rob Wiblin and Will MacAskill discuss MacAskill’s evolving views on long-termism, act/omission ethics, and the likelihood that we live in the most important period in history. The conversation covers his new papers on the “paralysis argument” and risk-averse altruism, critiques of the “hinge of history” hypothesis, and how these debates change EA strategy, cause prioritization, and views on existential risk, AI, war, and cultural lock-in.
Main Topics: MacAskill’s current work and book projects: MacAskill explains how he divides time between EA organizational work, GPI, and writing a new accessible-yet-academic book on the future and long-termism. He describes using speaking tours and audience feedback to refine his arguments. The paralysis argument and non-consequentialism: He and Andreas Mogensen argue that common non-consequentialist commitments—especially harm/benefit asymmetry plus the act/omission distinction—can imply that ordinary actions (like shopping or driving) are morally forbidden because they indirectly affect births, deaths, and harms over long causal chains. Long-termism under different moral views: The discussion distinguishes axiological vs deontic long-termism and explores whether non-consequentialists can preserve their views without collapsing toward consequentialism, including via ex ante Pareto reasoning or revised act/omission accounts. Risk aversion and altruism: MacAskill argues that altruistic risk aversion should concern world outcomes, not the agent’s personal sense of making a difference. He suggests risk aversion can still support extinction-risk reduction when the future is sufficiently valuable or uncertain. Are we living in the most influential time in history?: He defends skepticism toward the claim that this century is uniquely pivotal, arguing the prior should be low because the future may be extremely long, and evidence for a hinge-of-history moment is weaker than many assume. He revises his own estimate upward somewhat but remains cautious. Existential risk, AI, and other long-run threats: MacAskill updates his 100-year extinction-risk estimate downward and is more skeptical of specific AI doom narratives than before, while still seeing AI, war, bio-risk, totalitarianism, and value lock-in as serious long-term concerns. Implications for EA strategy and community culture: He argues EA should be more patient, research-heavy, friendly to other values, and less like a short-term startup. He emphasizes maintaining cooperative norms, cause diversity, and a long time-horizon for movement-building and influence.
Key Arguments: If non-consequentialists say causing harm is worse than allowing harm, then ordinary actions that alter future births and deaths can become massively harmful in expectation, producing a near-paralysis conclusion. Attempts to escape the paralysis argument by appealing to mediation through others, causal distance, or foreseeability do not clearly solve the problem; many intuitive cases still count as harms. A more promising non-consequentialist escape is ex ante Pareto reasoning, but that moves the view significantly toward consequentialism and may permit implausible actions such as killing one to save many if everyone ex ante prefers it. Altruistic risk aversion should be about the world’s welfare, not about whether the agent personally “made the difference”; on that basis, risk aversion can still favor extinction-risk reduction in some cases. The claim that we live at the most influential time ever starts from a very low prior because there may be many future centuries and a huge future population; the evidence so far is not strong enough to make the posterior very high. MacAskill’s estimate of existential risk in the next century has fallen sharply because he is less persuaded by the strongest AI-doom arguments and thinks many catastrophic scenarios are harder to realize than commonly assumed. Even if the present is not uniquely hingey, long-termist strategy may still favor saving resources, investing in research, and building durable institutions rather than maximizing immediate action. EA should prioritize cultural health—friendliness, intellectual honesty, low infighting, and broad cause flexibility—because the movement’s long-run influence may matter more than short-term growth.
Data Points: David Chalmers episode completion rate: 59% - Rob notes listeners who started the previous 4h42m episode on average finished 59% of it. Previous episode length: 4 hours 42 minutes - Rob cites this as the longest release yet. Book publication timing: late April - MacAskill’s book on moral uncertainty is said to be coming out in late April. Time to apply for GPI pre-doctoral research fellows: 12 hours after episode release - Rob warns the application window closes very soon after the episode airs. MacAskill’s talk feedback metric: 6 or 7 out of 7 - He measures how many audience members rate the talk as highly convincing after self-reporting low prior knowledge. Audience knowledge threshold: 4 or less out of 7 - MacAskill’s key metric focuses on people who came in with low prior familiarity. Reported convincingness: about 50% - He says roughly half of relevant audience members rate the talk highly convincing. Improved convincingness over time: maybe 60% - He suggests the metric may have risen somewhat during the speaking tour. MacAskill’s time allocation: about a quarter / quarter / bulk - He splits time between EA organizational work, GPI, and his book project, with the book taking most of it. Estimated 100-year existential risk (earlier view): 20% - He says his credence used to be about 20%. Estimated 100-year existential risk (current view): 1% - He says his credence has fallen to around 1%. Current plausible existential-risk interval: 0.1% to 1% - He says this is his current ballpark for century-scale existential risk. Probability of being at the most influential time: 0.1% to 1% - He says his updated estimate for the chance this is the hinge of history is in this interval. Outside-view prior suggested by Toby Ord: 5% to 10% - MacAskill reports Ord’s suggestion was much higher than his own prior. Expected ex ante risk aversion example: square of probability - He gives an example where extreme risk aversion might weight outcomes by p² rather than p. Horizon for discussion of current-era pivotality: next 50 years - He repeatedly frames some pivotal developments as potentially emerging over the coming half-century. Car crash mortality share: 1% to 2% - Used in the paralysis argument to illustrate that many deaths are contingent on ordinary actions. Toby Ord’s estimate for nuclear-war existential risk: 0.1% - MacAskill says Ord uses this as a rough figure in related discussions. Metaculus estimate of >95% population reduction: about 0.5% - MacAskill cites the forecasting platform’s aggregate estimate as broadly consistent with his low x-risk view.
Pivotal Quotes: "if you've got those two claims, then you've got to conclude that these perhaps millions of people... the fact that you've caused those deaths is worse than the corresponding amount of benefit" — Will MacAskill: Summarizing the core of the paralysis argument against non-consequentialist ethics. "the thing that I care about is just good stuff. The world goes well." — Will MacAskill: Explaining why altruistic risk aversion should focus on world outcomes rather than the agent’s personal sense of making a difference. "what we care about is our posteriors rather than priors" — Will MacAskill: His conclusion in the debate over whether this century is the most important in history.
Implications: Listeners should expect EA/longtermism to shift toward more research, more patience, and less certainty about AI-centric short-termism. The conversation argues for broader, more resilient cause prioritization and stronger cooperative norms in the movement.