Longtermism
Core insight: Future people matter morally as much as present people, and because there may be an enormous number of them, even small improvements to the long-run trajectory of civilization have greater expected moral value than almost any near-term intervention — making the deliberate, careful shaping of the long-run future among the most important moral priorities available.
How Each Book Addresses This
William MacAskill - What We Owe the Future — The Case for Making the Future a Moral Priority
MacAskill’s central contribution is building the philosophical scaffolding that makes the longtermist position coherent and actionable, not merely provocative. His argument operates through three premises that individually are nearly impossible to reject, but jointly generate substantial moral obligations:
-
Future people count morally as much as present people. There is no defensible principle by which temporal position alone makes a person less morally significant. We do not think geographic distance makes distant people morally less real. Temporal distance should not either.
-
There could be an enormous number of future people. If humanity survives for even 1% of the time mammals have existed, trillions of people will live after us. The potential scale of the future dwarfs the present.
-
We can make their lives better. Current decisions about institutions, technologies, values, and risks will substantially shape the conditions future people inherit. This is not speculative — it is the structure of civilizational history.
The implication: even small improvements in long-run trajectory have greater expected moral value than most near-term interventions, because the beneficiary population is so much larger.
The temporal discounting problem: MacAskill’s hiker thought experiment is the cleanest diagnostic for anti-longtermist intuitions. You drop a glass bottle while hiking; you know that if you leave it, a child will cut herself on the shards. Does it matter morally whether the child cuts herself next week, in a decade, or in a century? The intuitive answer is no — harm is harm whenever it occurs. This eliminates the most common implicit defense of near-termism: that temporal distance, like geographic distance, is morally irrelevant when it does not track genuine uncertainty about whether the harm will occur.
Two ways to improve the long-run future:
- Survival improvements: reducing extinction probability, ensuring humanity’s continued existence — increasing the quantity of future life
- Trajectory improvements: changing the quality, values, and institutions of civilization — increasing the quality of future life
These are distinct moral goals requiring distinct interventions. A civilization that averts extinction through totalitarian AI control hasn’t achieved a good long-run future — it has produced more of a bad one at vast scale.
The epistemic safeguard — the SPC framework: MacAskill is explicit that longtermism, unconstrained, can rationalize almost anything by appealing to speculative long-run effects. His safeguard is the Significance-Persistence-Contingency framework: an action merits longtermist priority only if it is (a) significant in scale, (b) persistent in effect, and (c) contingent in the sense that the outcome is sensitive to whether you specifically act. This prevents the framework from becoming a blank check for confident predictions about distant futures.
Career choice as the primary longtermist lever: For most individuals, career is the primary instrument of long-run impact — roughly 80,000 hours of productive output directed by one choice. MacAskill argues that career deserves moral deliberation proportionate to its importance: it is not a lifestyle preference but the highest-stakes moral decision most people will make.
How to apply:
- Apply longtermist criteria to major life decisions: does this work improve or worsen the long-run trajectory, and by how much relative to alternatives?
- Use the temporal discounting test when you notice yourself dismissing future effects: “Would I dismiss this if it happened to someone geographically distant instead of temporally distant?”
- Guard against longtermist overreach by applying the SPC filter to any proposed intervention — speculative long-run effects are not sufficient; the effect must be significant, persistent, and genuinely contingent on your action.
Isaac Asimov - Foundation Series — The Seldon Plan as Longtermist Design at Civilizational Scale
The Foundation Series is the vault’s most fully developed fictional instance of longtermist strategy. Hari Seldon’s foundational insight — that the Galactic Empire’s collapse is inevitable, that trying to prevent it wastes resources, and that the correct intervention is to reduce the subsequent Dark Age from 30,000 years to 1,000 years — is longtermism at its most structurally precise.
The Seldon Plan through MacAskill’s SPC framework:
- Significance: 29,000 years of Dark Age prevented — the largest-scale civilizational intervention Seldon can conceive
- Persistence: The Plan operates across a 1,000-year timeframe; the Second Foundation’s role is specifically to maintain the trajectory against perturbations
- Contingency: Without the two Foundations seeded at opposite ends of the galaxy, the Dark Age runs its full length; the Plan is genuinely contingent on the intervention
The Plan’s design principle — conditions over direct control: Seldon does not try to force the correct civilizational outcome through sustained power. He designs the Seldon Crises: situations at which the Foundation faces apparent existential threat, and the only viable solution is also the one that advances the Plan. Future actors are free to choose; Seldon has designed the conditions under which the right choice is always also the strategically rational one. This is the most sophisticated longtermist design tool: shaping the decision environment of actors centuries in the future rather than predicting or controlling what they will decide.
The Encyclopedia deception as longtermist tradeoff: The encyclopedists are recruited under false pretenses — they believe they are compiling the Encyclopedia Galactica; they are actually establishing the physical Foundation. Seldon’s calculation is that full transparency would expose the Plan to political destruction before it could begin. MacAskill’s SPC framework would evaluate this as legitimate precisely because the encyclopedists are not harmed (they live productive, meaningful scientific lives) and the deception is genuinely contingent on the Plan’s success.
The Second Foundation as redundancy design: Seldon built a Second Foundation because he recognized that any longtermist plan operating across 1,000 years will encounter events outside its design parameters (realized as the Mule). The redundancy — a mental-science Foundation complementing the physical-science Foundation — is the longtermist designer’s acknowledgment that no single-track plan is robust enough across centuries of unforeseeable perturbation. MacAskill’s SPC framework implicitly requires this: Significance and Persistence claims are only valid if the plan is robust to perturbation.
The Seldon Plan vs. MacAskill’s career-choice framing: MacAskill focuses on individual career choice as the primary longtermist lever. The Foundation adds a second lever: institutional design. The encyclopedists’ choice of career matters less than the institutional architecture Seldon designed — the two Foundations as compounding civilizational assets. At civilizational scale, institution-building (not just career-choosing) is the longtermist intervention that compounds across the relevant time horizon.
How to apply:
- Before dismissing any apparently failing institution or system, ask: is the correct intervention to prevent the failure, or to position the next system to succeed faster? Seldon’s acceptance of the Empire’s unsalvageability and redirection of energy to what comes next is the longtermist’s most counterintuitive productive move.
- Design longtermist plans for mediocre executors, not ideal ones. The Seldon Plan survives centuries of variable leadership because it creates conditions where the correct choice is always the rational one — independent of the moral quality of the actors making it.
Frank Herbert - Dune Series — The Golden Path: Voluntary Despotism as the Longtermist’s Extreme Case
Dune is the vault’s most extreme and most uncomfortable examination of longtermist commitment and its specific failure modes. The Golden Path — Leto II’s 3,500-year voluntary despotism — is the most demanding longtermist intervention in all fiction: it requires the planner to sacrifice their own humanity progressively over millennia, to be comprehensible to no one alive at any point in the 3,500 years, and to design for a dispersal whose benefits materialize only after the planner has been dead for centuries.
The Golden Path as longtermist diagnosis:
Leto II’s starting point is not a goal but a diagnosis: humanity is evolving toward monoculture — planetary-scale civilizational consolidation that makes species-level extinction possible through a single catastrophic event. His calculation shows that this trajectory leads to extinction with certainty over a long enough time horizon. The correct intervention is not stronger defenses (insufficient for long-horizon threats) but forcing dispersal so thorough that no single threat can reach all of humanity simultaneously.
MacAskill’s two categories of longtermist intervention appear explicitly in the Golden Path:
- Survival improvement: The Scattering prevents human extinction by making humanity impossible to eradicate completely
- Trajectory improvement: Prescience-immunity (Siona’s genetic line) ensures that no future prescient being can ever lock humanity into a single determined path — it improves the quality of all future civilizational development permanently
The epistemological problem of longtermist confidence:
Leto’s full prescient vision is the series’ most precise examination of the risk MacAskill’s SPC framework guards against: longtermist overconfidence. Leto can see the consequences of his plan across 3,500 years, but even perfect foresight doesn’t make the plan simple or painless — it requires carrying an unimaginably heavy burden, and the plan’s success depends on Leto correctly reading a probability distribution over millennia rather than a specific causal chain. His prescience does not eliminate uncertainty; it only makes him more confident about which trajectories close off which futures.
Paul Atreides as the longtermist failure case:
Paul sees the Golden Path and refuses to implement it. His refusal is not from ignorance — he understands what the plan requires — but from unwillingness to pay the personal cost: the progressive transformation of his body and ultimately his humanity. Herbert presents Paul’s refusal as the paradigmatic longtermist failure: seeing the long-term correct action clearly and declining to take it because the short-term cost is too high. Leto accepts the same burden Paul refused, with full understanding that his refusal condemned future generations to the extinction risk Paul saw but couldn’t face.
What Dune adds to MacAskill’s framework:
MacAskill’s longtermist argument centers on moral obligations to future people and career choice as the primary individual lever. Dune adds the psychological dimension: longtermist commitment requires not just believing that future people matter but the willingness to carry burdens whose beneficiaries will never know your name, whose costs are fully borne by you, and whose timeframe extends so far beyond your life that you will not see whether the plan worked. The 3,500-year timeframe is extreme; the psychological structure is the same at any longtermist scale. The Paul/Leto contrast is the sharpest examination in any book of what genuine longtermist commitment requires beyond philosophical agreement.
How to apply:
- Apply the Golden Path test to any significant organizational consolidation: if a single threat could eliminate the entire consolidated entity, the correct response is dispersal, not stronger defenses. Resilience through diversity beats resilience through strength at long time horizons.
- The Paul/Leto diagnostic: distinguish between longtermist agreement (knowing the long-run correct action) and longtermist commitment (being willing to pay the full cost of executing it). Most longtermist failure happens in this gap.
Max Tegmark - Life 3.0 — The Cosmic Endowment Argument: Longtermism at Maximum Scale
Tegmark provides the strongest cosmological grounding for the longtermist position in the vault: the Cosmic Endowment Argument transforms AI safety from a local human concern to the most consequential intervention available at cosmic time horizons.
The argument:
The observable universe contains roughly 10²³ stars, each potentially hosting planets. If life originating on Earth eventually colonizes even a small fraction of this cosmic endowment, the total amount of experience and flourishing made possible is astronomically large compared to anything that happens in the near-term on Earth alone. Conversely, if a poorly aligned AGI eliminates Earth-originating life or permanently locks in suboptimal values, it destroys this cosmic potential. The asymmetry between the upside (enormous positive future) and the downside (permanent loss of cosmic potential) makes AI safety one of the highest-expected-value interventions available by a wide margin.
Why this is more precise than MacAskill’s framing:
MacAskill’s longtermist case is built on the moral weight of future people and the scale of potential future generations. Tegmark’s Cosmic Endowment Argument provides cosmological grounding for the scale claim: not just “many future people may exist” (a trajectory extrapolation) but “the physical resources of the observable universe represent an upper bound on future experience and flourishing” (a physics argument). The cosmic endowment is not a speculative prediction — it is a fact about the observable universe. Whether life from Earth can access it depends on successfully navigating the AI transition.
Connection to MacAskill’s SPC framework:
Tegmark’s AI safety work passes MacAskill’s SPC filter with maximum scores:
- Significance: Cosmic-scale — the entire cosmic endowment of 10²³ stellar systems is at stake
- Persistence: Permanent — a misaligned AGI that eliminates or permanently constrains Earth-originating life forecloses the cosmic endowment forever; a successfully aligned AGI opens it indefinitely
- Contingency: Highly contingent — current decisions about AI safety research funding, governance frameworks, and development practices are genuinely path-dependent; outcomes differ substantially based on current choices
The two-planet thought experiment:
Tegmark’s epilogue presents Planet A (developing AGI with strong safety culture, good governance, broadly distributed values) and Planet B (developing AGI in a competitive race with safety neglected). Both achieve AGI roughly simultaneously. The long-run outcomes diverge massively. The critical decisions happen early — before the technology arrives, when safety culture, institutions, and governance frameworks are established. This is the longtermist’s practical argument: the expected-value calculation changes dramatically based on early-stage decisions about safety and governance, not just capability.
The Dune parallel:
Leto II’s 3,500-year plan specifically addresses the cosmic-endowment logic at a smaller scale: humanity’s geographic concentration makes it vulnerable to a single catastrophic event. The Scattering prevents this. Tegmark’s argument is structurally identical at cosmic scale: Earth’s concentration of intelligent life makes it vulnerable to a single catastrophic AGI event. Successfully navigating the AI transition is the “Scattering” — the intervention that makes the cosmic endowment accessible rather than permanently foreclosed.
How to apply:
- Use the Cosmic Endowment framing to evaluate the relative importance of different AI safety interventions: interventions that address existential risk or permanent value lock-in have dramatically higher expected value than interventions that address serious-but-recoverable near-term harms.
- Apply MacAskill’s career-choice lens to the AI safety domain: work on AI safety represents one of the clearest cases where the SPC test gives maximum scores — significance (cosmic), persistence (permanent), contingency (high).
- The two-planet thought experiment as a diagnostic for any AI policy: “Would this decision help Planet A or Planet B?” — not “does this produce near-term capability advantages?”
Nick Bostrom - Superintelligence — The Pre-Transition Window: Maximum Leverage, Minimum Time
Bostrom’s Superintelligence provides the most operationally specific treatment in the vault of the intervention window argument — why the pre-superintelligence period is the highest-leverage moment for longtermist action, and how the control problem window differs from MacAskill’s general plasticity argument.
The pre-transition window as the tightest plastic moment:
MacAskill’s plasticity argument: civilizational values are currently plastic, and this window will close when transformative AI or global power consolidation hardens the system. Bostrom’s argument is more precise and more urgent: the control problem window specifically closes at the moment of the first decisive strategic advantage, which may be a fast takeoff event offering very limited warning. Unlike MacAskill’s gradual lock-in concern (which could occur over decades), Bostrom’s window may close in months or years once the capability threshold is crossed.
This creates a hierarchical urgency: MacAskill’s general longtermist case (act during the plastic period) is subsumed under Bostrom’s more specific claim (the AI safety window may close discontinuously faster than general institutional windows). A longtermist who acts on MacAskill’s timeline but not Bostrom’s urgency may act too late for the specific intervention that matters most.
The expected value argument under uncertainty:
Even under substantial uncertainty about the probability of fast takeoff or decisive strategic advantage within the investment window, the expected value of control problem research is extreme. The calculation: a small probability of the transition occurring soon, multiplied by the permanent nature of the outcome (a misaligned singleton forecloses the entire cosmic endowment of 10²³ stellar systems), produces an extremely large expected value for marginal investment in alignment. This is MacAskill’s SPC framework at maximum compression: Significance (cosmic endowment), Persistence (permanent), Contingency (the alignment research community is small enough that marginal additions produce meaningful change).
The differential progress argument as longtermist lever:
Bostrom’s most practically actionable longtermist claim: it is possible to accelerate safety-relevant research faster than overall capability research — differential technological development. Investment directed at differential progress has higher expected value than either general capability investment (which advances the transition timeline without improving safety) or general coordination effort (which is slow relative to capability timelines). The longtermist career choice that passes the SPC test is specifically the research direction that improves the ratio of safety understanding to capability development.
How to apply:
- Apply the intervention window test to any longtermist career or investment choice: does this work improve safety understanding relative to capability development (differential progress), or does it advance capability without proportionate safety improvement?
- SPC analysis of control problem research: Significance (cosmic endowment at stake), Persistence (permanent — a solved control problem allows access to all future value; an unsolved one forecloses it permanently), Contingency (highly path-dependent — the research community is small enough that marginal contributions produce meaningful change).
- The fast takeoff urgency test: don’t assume that the window for intervention has the same timeline as MacAskill’s general plasticity window. If fast takeoff is possible, the relevant window may be shorter and less legible than gradual lock-in processes.
Cross-Book Pattern
| Book | The Longtermist Claim | Mechanism |
|---|---|---|
| Nick Bostrom - Superintelligence | The pre-transition window as the tightest plastic moment: the control problem window closes at the moment of decisive strategic advantage — potentially a fast takeoff event, not a gradual process, giving much less time than general lock-in timelines; expected value of alignment work is extreme under any non-negligible probability of near-term transition; differential technological development (accelerating safety faster than capability) as the highest-SPC-score longtermist career direction | The fast takeoff urgency test: the relevant intervention window may close discontinuously faster than MacAskill’s general plasticity window — longtermist timing calibrated to general institutional timelines may be systematically too slow for the AI-specific transition; differential progress as the specific mechanism by which current investment shifts the probability distribution of the transition going well |
| William MacAskill - What We Owe the Future | Explicit philosophical case: future people count, many may exist, we can affect their lives — therefore improving the long-run trajectory is a key moral priority | SPC framework as the constraint on legitimate longtermist action; career choice as the primary individual lever |
| Isaac Asimov - Foundation Series | The Seldon Plan: accept civilizational collapse as inevitable, reduce the subsequent Dark Age from 30,000 years to 1,000 years through two strategically placed Foundations | Conditions-over-control design: crises structured so the correct choice is always strategically rational for future actors; Second Foundation as redundancy against unpredictable perturbation; Encyclopedia deception as the longtermist tradeoff between transparency and plan viability |
| Frank Herbert - Dune Series | The Golden Path: Leto II’s 3,500-year voluntary despotism to prevent human extinction through forced dispersal and prescience-immunity; Paul Atreides as the longtermist failure case (sees the plan, refuses to pay its cost) | MacAskill’s two categories operationalized: the Scattering (survival improvement) + Siona’s genetic line (trajectory improvement — permanent prescience-immunity); Paul/Leto contrast as the sharpest fictional examination of the gap between longtermist agreement and longtermist commitment |
| Max Tegmark - Life 3.0 | The Cosmic Endowment Argument: 10²³ observable-universe stars represent the upper bound on future flourishing; the AGI transition is the primary decision point determining whether Earth-originating life accesses or permanently forecloses this endowment; the two-planet thought experiment as the practical longtermist design frame | MacAskill’s two categories explicitly instantiated: survival improvement (preventing AGI-enabled extinction) + trajectory improvement (aligning AI values to preserve moral plasticity and cosmic optionality); SPC maximum scores — significance (cosmic), persistence (permanent), contingency (high); critical decisions happen before the technology arrives, when safety culture and governance frameworks are established |
| Bill Gates - How to Avoid a Climate Disaster | Climate change as the canonical longtermism case: emissions made now (and Green Premium reduction work done now) compound across all future people through atmospheric concentration that persists for centuries; the temporal asymmetry — we emit, future people bear the cost — is the moral structure of the climate problem | Long-horizon premium-reduction investment has compounding moral leverage across generations; Innovation Plus Deployment as the two-track strategy operationalizes longtermism for climate — deploy mature technologies now (immediate emissions reduction) AND fund breakthrough R&D for unsolved sectors (compound future impact); the 51-to-0 framing is longtermism applied to atmospheric concentration: only zero stops the accumulation, partial reduction merely slows continued damage to all future people |
Related Concepts
- Concept - Responsibility & Meaning — Longtermism reframes the scale at which voluntarily accepted burden can be carried; obligations to future people are the largest-scale version of voluntary meaningful burden
- Concept - Value Lock-In — The specific failure mode longtermism is most urgently trying to prevent: values becoming permanently entrenched before they are wise
- Concept - Moral Circle Expansion — Longtermism as the next step in the historical pattern of moral circle expansion — now extending concern to people in future times
- Concept - The Great Filter — Extinction risk intersects directly with longtermism; both are concerned with what forecloses the long-run future
- Concept - Big Bets & Calculated Risk — Career choice and cause prioritization under the SPC framework are the longtermist’s primary calculated bets