MacAskill, William

philosopher; co-founder of the effective altruism movement; co-founder of Giving What We Can, 80,000 Hours, and the Centre for Effective Altruism

William MacAskill: Effective Altruism

Lex Fridman Podcast

• William MacAskill
• effective altruism
• longtermism
• existential risk
• global poverty
• animal welfare
• GiveWell
• AI safety
• moral philosophy
• Peter Singer

Core ideas

MacAskill defines effective altruism (EA) as the project of using evidence and reason to determine how to do the most good with one's limited time and money, then acting on those conclusions — distinguishing it from ordinary charitable impulses by its insistence on the same analytical rigor people already apply to, say, evaluating a business investment. He opens with his vision of a "long reflection": a hoped-for future period of post-scarcity stability in which humanity can spend an extended time (potentially tens of thousands of years) deliberating on ethics before taking irreversible, civilization-shaping actions like interstellar expansion, and expresses concern that current media formats (he singles out Twitter's brevity and virality-driven incentives) are poorly suited to the kind of rational, nuanced discourse such a reflection would require, contrasting this with long-form podcast conversation as closer to humans' evolved communication style.

On effective giving, MacAskill explains that most people's charitable impulses lack the rigor applied to other consequential decisions, and describes his organizations' approach: identifying cause areas by scale, tractability, and neglectedness, then within those areas funding evaluators like GiveWell that rank interventions by measurable cost-effectiveness (citing insecticide-treated bed nets, roughly $5 each and highly effective against malaria, and cash transfers via GiveDirectly as baseline comparisons). He grounds the case for global-poverty-focused giving in a "diminishing marginal utility of money" argument: since a given proportional income increase yields roughly equal well-being gains regardless of starting wealth, and since rich-country citizens are roughly 100 times wealthier than the extreme global poor, a dollar donated to the poorest is roughly 100 times more impactful than a dollar spent on oneself. He personally donates about 20% of his income above a self-set bar, describes redirecting his academic career from abstract philosophy of logic toward applied ethics as a direct consequence of taking EA's reasoning seriously, and traces the movement's founding to two converging strands: his and Toby Ord's Oxford-based "Giving What We Can" (inspired by Peter Singer's drowning-child argument for a moral obligation to aid distant strangers) and Holden Karnofsky and Elie Hassenfeld's GiveWell, founded by two former hedge-fund analysts dissatisfied with existing charity evaluation.

MacAskill distinguishes EA from both Objectivism (which he says is fundamentally opposed to EA if it means literal ethical egoism, and empirically implausible if it claims self-interest coincidentally maximizes social benefit) and unconstrained free-market "just do what you're good at" reasoning, arguing markets systematically under-represent the interests of parties who can't participate in them — non-human animals, future generations, and (to a lesser extent) the very poor. Addressing criticism from both political directions — from the left, that EA doesn't sufficiently challenge capitalism as a root cause; from some on the right (in reaction to his Joe Rogan appearance), that altruism itself is somehow "unmanly" — he responds that EA's emphasis on individual agency and consciously choosing one's own priorities over socially default ones is, in his view, actually compatible with values of self-determination and control that attract people to more individualist philosophies. On animal welfare, he states his best guess is that consciousness is a graded, information-processing-based phenomenon (not a binary "on/off" property), such that animals plausibly experience suffering to varying degrees correlated loosely with neural complexity, and identifies factory farming (roughly 50 billion animals slaughtered annually, most under practices he says would be considered inhumane by nearly any ethical standard) as the highest-leverage animal welfare cause, citing corporate "cage-free" pledge campaigns as currently more tractable than direct regulatory reform given agricultural-industry regulatory capture in the U.S. He also notes, while affirming this doesn't diminish the priority of factory farming, that wild-animal suffering (starvation, disease, and predation, especially under r-selected reproductive strategies producing huge numbers of offspring of which few survive) is arguably even more severe and underexamined.

On existential risk, MacAskill grounds his concern in a "longtermist" argument: given that civilization could plausibly persist for millions or billions of years, the sheer number of potential future people vastly outweighs the present generation, making permanent civilizational derailment (extinction or irreversible "lock-in" of bad values) an enormous moral stake. He discusses nuclear weapons as a first proof-of-concept that humanity could destroy itself, citing JFK's own estimate of a 1-in-3 chance of nuclear exchange during the Cuban Missile Crisis, and argues that mutually-assured-destruction deterrence depends on fissile material being hard to acquire and control — a property future biotechnology (e.g., hypothetical home DNA-synthesis capability) may not share, especially against actors (he cites the 1990s Aum Shinrikyo cult) not deterred by their own deaths. On AI risk, he distinguishes the "classic" fast-takeoff scenario (a single system recursively self-improving into a dominant, potentially misaligned superintelligence, associated with Nick Bostrom and Eliezer Yudkowsky) from what he considers a more probable danger: AI-driven economic and technological acceleration concentrating power in a small number of actors who could then "lock in" their own values indefinitely via AI systems that, unlike human rulers, don't die of old age — a scenario he frames as bad not because any particular locked-in value set is evil, but because locking in any single value set forecloses the ongoing moral exploration (the "long reflection") he considers essential to eventually discovering better values. He outlines three ways individuals can act on AI risk (technical AI safety research, AI policy work, and donating to organizations like the Long-Term Future Fund, the Center for Human-Compatible AI, or the Future of Humanity Institute), states he considers a broadly good outcome the most likely trajectory but that even small probabilities of civilizational-scale catastrophe warrant serious effort, and largely agrees with Elon Musk's public concern about existential risk while noting the exact probability estimate matters less than taking the stakes seriously. Closing, he says the one thing he would most want to change about the world is that people generally don't care about, or think about, the welfare of future generations.

Excerpts