Dwarkesh PodcastWill MacAskill - Longtermism, Effective Altruism, History, & Technology
CHAPTERS
- 0:00 – 1:19
Moral progress isn’t over: resisting value lock-in and enabling a “long reflection”
MacAskill opens by arguing that today’s dominant values are historically contingent and shouldn’t be treated as the endpoint of moral progress. He introduces the idea of a “long reflection”: creating conditions where reasoned debate and empathy can steer future values before irreversible lock-in occurs.
- •Western values could have been very different under alternate historical outcomes (e.g., Nazis winning WWII)
- •We should prioritize discovering what’s actually morally right, not just preserving whichever values won historically
- •Long reflection: delay irreversible decisions so moral inquiry can do more work
- •Factory farming as an example of contingent moral catastrophe
- •Big-picture moral questions often fall through disciplinary cracks
- 1:19 – 2:44
Why Effective Altruism took off (this time): internet coordination, luck, and prior precedents
Dwarkesh asks why Effective Altruism (EA) succeeded when similar ideas appeared earlier in history. MacAskill frames EA’s rise as partly contingent—shaped by enabling technologies, network effects, and fortunate personal connections—rather than inevitable.
- •Historical precursors: Mohists, early utilitarians, Peter Singer’s 1970s arguments
- •EA’s uptake was delayed despite strong underlying ideas
- •The internet helped like-minded people find each other and coordinate
- •Catalytic contingencies: key meetings and early organizational formation
- •EA as a ‘good idea waiting to happen,’ but timing mattered
- 2:44 – 6:01
Contingent values and the case against “locking in” modern Western morality
They explore whether contingency implies our current values are probably mediocre, and whether we should distrust our own moral confidence. MacAskill agrees contingency should humble us, while still arguing we may be above historical average due to fragile-but-real moral gains.
- •If history had gone differently, people might sincerely endorse abhorrent norms
- •Contingency is evidence we’re far from moral truth and should remain open to progress
- •Counterfactual: an Industrial Revolution elsewhere might have avoided factory farming
- •Reasons to think we’ve made genuine gains (abolitionism, feminism, liberal democracy)
- •Long reflection aims to strengthen the ‘force’ of moral argument rather than treat progress as random
- 6:01 – 8:42
First principles vs tradition in a fast-changing world (and what COVID taught forecasters)
Dwarkesh probes whether longtermism should be more Burkean—leaning on tradition and historical wisdom. MacAskill argues rapid technological and social change makes inherited heuristics less reliable, and cites early pandemic seriousness in EA circles as a win for updating on evidence.
- •Tradition-following works better in low-change environments; modernity is high-change
- •Humans are culturally conformist by default—sometimes to our detriment today
- •EA could benefit from more historical literacy, even while doing first-principles work
- •COVID as an example: early alarm versus institutional inertia
- •The marginal value of history: under-supplied in EA compared to philosophy/econ
- 8:42 – 12:02
How contingent is technology and growth? Recoveries, incentives, and ‘near-inevitability’ post–Industrial Revolution
They discuss whether shocks like wars and plagues matter long-run for economic growth, versus moral progress. MacAskill claims technological/economic progress is driven by strong incentives and tends to re-emerge, with contingency mostly affecting timing—especially after industrialization.
- •Economic recovery after catastrophe suggests strong pull toward growth trajectories
- •Examples of early/latent tech: steam-engine-like devices, semaphore, flying shuttle
- •Pre–Industrial Revolution: larger timing contingency (centuries to millennia)
- •Post–Industrial Revolution: many technologies likely arise within decades regardless
- •Structural argument: once one society grows faster, it eventually dominates globally
- 12:02 – 15:44
Who changes history most? Counterfactual impact of scientists, philosophers, activists, and dictators
Dwarkesh asks which careers have the highest counterfactual impact if many technologies are ‘inevitable.’ MacAskill argues individuals can matter most through ideas, ideology, and moral movements—especially under dictatorship—while even great scientists may only shift timelines by decades.
- •Green Revolution: Borlaug likely accelerated life-saving tech rather than uniquely enabling it
- •Einstein example: some breakthroughs (general relativity) plausibly decades ahead of time
- •Long-run contingency often runs through ideology: Marx/Engels, religious founders, abolitionists
- •Dictatorships amplify individual contingency (leader values strongly shape outcomes)
- •Technology can ‘cement’ ideology, extending its persistence via surveillance/biotech/AI
- 15:44 – 18:57
Stagnation worries: fertility, research productivity, and why talent clusters ‘punch above weight’ but don’t dominate history
They turn to MacAskill’s stagnation concerns: progress depends on population, R&D intensity, and productivity. Dwarkesh raises counterexamples like Bell Labs; MacAskill grants organizational effects but argues they’re small compared to population-scale and culture-wide incentives over centuries.
- •Progress model: population × share devoted to R&D (and productivity)
- •Fertility decline + falling research productivity as long-run headwinds
- •Historical clusters (Athens, Venice, Baghdad, Bell Labs) reflect cultural incentive differences
- •Talent can be redirected into non-compounding pursuits (e.g., theology, music)
- •Organizational excellence matters, but broad culture and scale dominate in the ‘grand sweep’
- 18:57 – 24:08
Longtermist institutional reform: futures assemblies, incentives, and the futarchy temptation
Dwarkesh asks about MacAskill’s proposals for embedding long-term thinking into governance. MacAskill is cautious: future people can’t lobby, so institutions are vulnerable to co-option; still, he sketches mechanisms like sortition, delayed evaluation, and multi-metric forecasting indices.
- •Core problem: representing future interests is structurally hard (no constituency)
- •Proposal: randomly selected ‘future assembly’ with some legal power
- •Incentives via delayed assessment (30-year lookbacks) and chained accountability
- •Metrics could include GDP, homelessness, tech progress, and catastrophe-risk indices
- •Futarchy: ‘vote on values, bet on beliefs’—promising in theory, hard in practice (liquidity/gaming)
- 24:08 – 26:52
When reforms get gamed: environmental review as a cautionary tale and a pivot to narrower long-term wins
Dwarkesh draws an analogy to environmental impact reviews being exploited to block projects. MacAskill agrees the risk of co-option is potentially devastating and admits uncertainty about better design, suggesting longtermists may do better focusing on targeted policies that help the long term without pretending to ‘represent the future.’
- •Institutions designed to protect diffuse interests can be strategically exploited
- •Environmental review boards as an example of mechanism gaming and mission drift
- •MacAskill is unsure whether failures are intrinsic or design-contingent
- •Shift in emphasis: prioritize narrow, robust reforms (e.g., bio lab liability insurance)
- •Practical skepticism about ambitious constitutional-level longtermist bodies
- 26:52 – 29:52
Are companies longtermist? Institutional half-lives, universities, religions, and the danger of lock-in
They examine whether corporations’ discounted cashflow incentives make them longtermist. MacAskill argues companies are short-lived and leaders rarely optimize for institutional longevity, while universities and religions show much longer time horizons—raising both benefits and lock-in risks.
- •Corporate ‘half-life’ is surprisingly short (order of a decade)
- •Universities and religions can persist for centuries or millennia, shaping long-run norms
- •Longevity changes what decisions institutions even consider (e.g., 400-year traditions)
- •Lock-in can be good or catastrophic depending on what’s being locked in
- •US Constitution as a ‘plasticity moment’: remarkable success with near-miss catastrophic clauses
- 29:52 – 35:48
Living in an era of plasticity: global integration, AI-enabled ideological lock-in, and the race before space settlement
MacAskill explains why the current era may be unusually malleable: global communication, moral diversity, and rapid change. The danger is that plasticity could end if a world government plus AI (or digital rulers) enables stable ideological control before humanity spreads into space.
- •Plasticity driver #1: unprecedented global connectivity + diversity of moral views
- •Plasticity driver #2: plausible end to rapid moral change via monoculture/ideological conformity
- •AI as a key lock-in technology: digital rulers could be durable and hard to dislodge
- •Space settlement could preserve long-run diversity—but control might arrive first
- •MacAskill puts nontrivial probability on near-term control scenarios (e.g., ~10%)
- 35:48 – 40:14
How good can the future be? Subsistence ems, happiness evidence, and whether most lives are worth living
They debate Robin Hanson-style scenarios where digital minds (‘ems’) end up at subsistence due to competition. MacAskill argues subsistence doesn’t imply misery, then discusses survey evidence suggesting most people judge their lives positively—even across rich and poorer contexts—while cautioning about sampling and cultural confounds.
- •Malthusian/subsistence equilibrium doesn’t determine subjective wellbeing
- •Ems could be resource-poor yet extremely happy (in principle)
- •Commissioned survey: similar (low) shares in US/India report lives as net negative
- •Possible confounds: sampling differences, religion/reincarnation beliefs, generalizability limits
- •Additional method: ‘time you’d rather skip’ studies suggest ~10% experience net-negative days
- 40:14 – 50:55
Contra ‘just boost growth’: predictability, targeted risk reduction, and disputes with Cowen/Collison-style skepticism
Dwarkesh asks whether longtermism risks over-optimizing for what seems salient today, missing truly important but currently invisible levers. MacAskill engages a rival strategy—focus on historically robust goods like growth—and argues some technological trends and risks are predictable enough to justify targeted work on AI governance and catastrophic bio/war risks.
- •Critique: long-run impacts are hard to foresee; maybe it’s a ‘mug’s game’ to try
- •MacAskill’s reply: some trends are highly predictable (e.g., Moore’s Law-like trajectories)
- •EA/forecasters taking COVID and pandemics seriously early as evidence of usable foresight
- •Targeted priorities: AI safety/governance, preventing engineered pandemics, avoiding major war, preventing value lock-in
- •Value change can be faster today (EA, gay rights) though timelines still matter if AGI is soon
- 50:55 – 57:02
Why academia avoids big-picture questions (and MacAskill’s bet on building institutions)
In the closing stretch, they touch on MacAskill’s career choice, a humorous aside, and then return to a serious theme: many crucial long-horizon questions are poorly incentivized in both markets and academia. MacAskill floats the possibility of creating a new kind of university to do education and research better, then wraps with book and community resources.
- •MacAskill stays in academia because prestige/platform helps him develop and spread ideas
- •Big-picture topics are neglected: too grand, interdisciplinary, and speculative for standard incentives
- •Academia shifted toward incrementalism; older philosophy rewarded grand syntheses
- •Interest in founding a new university to improve education and research quality
- •Outro: book release details, Giving What We Can, 80,000 Hours, and closing credits
