William MacAskill on Effective Altruism, Moral Progress, and Cultural Innovation

William MacAskill — Oxford philosopher and co-founder of the effective altruism movement — joins Tyler Cowen in Ep. 156 to defend longtermism, wrestle with the limits of expected-value reasoning, and argue that EA itself is best understood as a form of cultural innovation rather than an optimisation exercise.

Key ideas

1. Moral uncertainty demands a portfolio of views, not a single theory. MacAskill rejects both strict utilitarianism and first-order moral pluralism. He argues instead for meta moral uncertainty — holding calibrated credence across multiple moral frameworks and acting on their weighted average. This is like a judge who acknowledges she could be wrong about the law and weighs competing interpretations before ruling, rather than treating one code as infallible.
2. Almost all expected value lies in the long-run future. If humanity flourishes for millions of years, the number of people yet to live dwarfs those alive today in the same way that most of a book’s pages lie ahead of the first chapter. Reducing existential risk — extinction, civilisational collapse — is therefore the highest-leverage intervention, far outweighing improvements to current welfare at the margins.
3. Expected-value theory breaks down at the extremes. MacAskill concedes he will not take a 51/49 double-or-nothing bet on civilisation, nor follow infinite-ethics arguments to their logical conclusions. He locates the failure specifically in multiplying tiny probabilities by enormous values — Pascal’s wager territory — rather than diagnosing a general weakness of consequentialist reasoning. He and Cowen disagree on where exactly to get off what MacAskill calls ‘the train to crazy town’.
4. Culture is the deeper variable — and EA is itself cultural innovation. MacAskill agrees with Cowen’s critique that EA has over-indexed on optimisation and measurability. His reframe: the scientific revolution was primarily a cultural shift — people started experimenting, tinkering, trusting evidence — and EA could matter in the same way by propagating a culture of impartial concern, accurate beliefs, and careful reasoning, regardless of any particular intervention it funds.
5. Hinge-yness is real but epistemically fraught. The current period is unusually influential — technological change that has compounded at roughly 2 per cent per year for 200 years cannot continue for 10,000 without consuming more atoms than exist within 10,000 light-years. We are in a narrow, anomalous window. But MacAskill grants that people in every era have felt their moment uniquely pivotal, and that predicting which levers matter most — foreign policy, AI governance, institutional design — remains extraordinarily hard.

Content

Moral uncertainty and the structure of ethical reasoning

MacAskill opens by distinguishing his position from both utilitarianism (one source of value) and first-order pluralism (many irreducibly competing values). His view is meta pluralism: we should hold credence across many moral frameworks — utilitarian, deontological, virtue-ethical — and take a weighted average of their verdicts. This mirrors how a careful judge treats competing lines of precedent: she doesn’t simply declare one line correct and ignore the others; she weighs them in proportion to their persuasiveness.

This structure explains why MacAskill resists Tyler Cowen’s repeated invitation to simply cap the domain of consequentialist reasoning at some comfortable scale. MacAskill argues that setting such a cap is itself an arbitrary choice: nothing qualitatively different happens when the numbers go from a thousand lives to a billion. The better move is to identify where the maths itself goes wrong — specifically, in operations involving infinities or near-zero probabilities multiplied by near-infinite values — and quarantine the failure there, rather than abandoning scale-sensitive reasoning across the board.

The repugnant conclusion and population ethics

Population ethics is the branch of ethics that asks how we should weigh bringing new people into existence, or preventing their existence, against improving the lives of people who already exist. The repugnant conclusion — formally proved by philosopher Derek Parfit — is that any consistent set of values implies that a vast population of people with lives barely worth living is better than a small population of people living in genuine bliss, provided the aggregate well-being is higher.

MacAskill’s position is that the repugnant conclusion, while counterintuitive, is actually the least bad bullet to bite. Every moral view must take a stance on population ethics; the paradoxes afflict deontologists and virtue ethicists as much as consequentialists. He therefore resists Cowen’s framing — that we should simply rule certain problems out of bounds — and insists on confronting them even at the cost of uncomfortable conclusions.

Expected value and where the reasoning breaks

Expected value is the tool borrowed from decision theory: multiply each possible outcome by its probability, sum the results, and choose the action with the highest total. It underlies most EA cost-effectiveness analysis (e.g. a bed net costing $4,500 per life saved versus a deworming programme costing $100 per life saved).

MacAskill and Cowen replay the well-known stress tests. Pascal’s wager: a tiny probability of infinite divine reward dominates any finite calculation. The St Petersburg paradox: a fair coin tossed until it lands heads, with payoff doubling each round, has infinite expected value but no sane person would pay much for it. MacAskill’s response is surgical: the pathology lies specifically in the interaction of extreme probabilities with extreme values, not in the expected-value framework per se. He explicitly declines a 51/49 gamble on civilisation, and acknowledges that the 90/10 version, played 200 times, creates the same problem.

Longtermism: why the future dominates the ledger

Longtermism is the view that improving the long-run future is the most pressing moral priority, given how many people could exist if humanity flourishes for millions of years. MacAskill offers a concrete illustration: at 2 per cent annual economic growth sustained for 10,000 years, humanity would produce economic output equivalent to roughly a trillion trillion current civilisations — more than one for every atom within 10,000 light-years. That plainly cannot happen, which means we are living through a temporary, anomalous spike in the rate of change. This is the basis for thinking we are in an unusually ‘hinge-y’ moment.

The practical implication is that reducing existential risk — the probability of extinction or permanent civilisational collapse — has extremely high expected value, because it preserves access to all that future value. Even a tiny reduction in extinction risk is worth more than large improvements to current welfare.

MacAskill adds a complication Cowen presses on: if we might be living in a simulation, all long-run bets are shakier. Under causal decision theory (the view that only the causal consequences of your actions matter), simulation shrinks the expected future; under evidential decision theory (the view that your actions are also evidence about how relevantly similar agents act), it matters less. MacAskill prefers causal decision theory and concludes: in a simulated world, near-term actions become relatively more attractive.

The culture critique and EA as cultural innovation

Cowen’s most pointed challenge is sociological: EA discussions — online and in person — drift relentlessly toward optimisation and measurement. How many chickens is a cow worth? Which deworming programme per dollar? This mindset, Cowen argues, systematically underweights the role of culture, institutions, and the long diffuse influence of things like opera, civic ritual, and classical ideals on the quality of a civilisation.

MacAskill’s response surprises him by conceding the point. He agrees that EA’s demographic skew toward STEM fields and high educational performance has produced a blind spot for ‘mushy’ variables — institutions, norms, trust, aesthetic tradition. He then reframes: the scientific revolution was not primarily a set of discoveries but a cultural shift — a new habit of experimenting, tinkering, and trusting empirical evidence over authority. He sees EA as potentially a similar cultural innovation: propagating a subculture of impartiality, altruistic motivation, and epistemic rigour. Its long-run impact may lie less in any specific grant it funds than in the kind of people and reasoning it normalises.

He credits Joseph Henrich’s work on cultural evolution — the view that humans are best understood as homo culturalis, creatures whose behaviour is governed by shared beliefs about how the world should be, not just by material incentives — as having substantially changed his thinking.

The political economy of EA and Conquest’s Second Law

Cowen raises Conquest’s Second Law: any institution not explicitly designed to hold a particular political position will drift left over time, as seen in the Ford Foundation, Rockefeller Foundation, and Pew Charitable Trusts. MacAskill acknowledges the pattern without fully explaining it. He notes that EA’s existing political skew — socially liberal, economically moderate — was not deliberately engineered but emerged from the selection effects of recruiting people who care about cost-effective global health interventions. He expects inertia to keep EA roughly in that range but concedes uncertainty about the underlying mechanism driving the historical pattern.

Related

• William MacAskill — guest; Oxford philosopher and co-founder of the effective altruism movement
• Tyler Cowen — host
• Deciding Under Uncertainty — theme; MacAskill’s account of moral uncertainty as a weighted portfolio of views bears directly on decision-making under normative uncertainty