---
title: "The Last Step"
subtitle: "What the strongest case for better futures is missing"
author: Elias Kunnas
description: "MacAskill's Better Futures is the most sophisticated case for caring about civilizational trajectories. It gestures toward systemic value, preserving possibility-space, and complex systems as intrinsically worth sustaining. It stops one step short of a foundation that would make all of this robust. Here's the step."
canonical: https://kunnas.com/articles/the-last-step
url: https://kunnas.com/articles/the-last-step.md
date_published: 2026-02-18
date_modified: 2026-02-19
corpus_frame_url: https://kunnas.com/articles/how-to-read-this.md
---
## How to read this corpus

The corpus applies one lens to many domains: what mechanisms produce the outcome? It shares four methodological commitments and one explicit directional commitment. Each linked page argues for its part; the links are derivations and disputes, not evidence inherited by every page. The directional commitment does not by itself settle system boundary, distribution, sacrifice, or institutional authority.

1. **Mechanisms are what act.** Incentive gradients, selection pressures, feedback loops, and capital stocks produce the distribution of outcomes. Intentions, labels, official categories, and stated values are evidence about mechanisms, or are themselves coordination mechanisms. They are not causal substitutes. — [Mechanism Realism](https://kunnas.com/articles/mechanism-realism.md) · [Only Selection](https://kunnas.com/articles/only-selection.md)
2. **The reference telos is sustained flourishing.** The broadest achievable adaptive safety margin over deep time — not the continuity of any incumbent state, coalition, institution, or doctrine. A mechanism's own stated goal can still serve as a local proof obligation — showing that its incentives defeat even the purpose it claims is a bounded finding — but meeting that goal establishes nothing about the margin. — [Flourishing Is Maximum Safety Margin](https://kunnas.com/articles/flourishing-is-maximum-safety-margin.md)
3. **Law, rights, legitimacy, democracy, markets, and sovereignty are mechanisms under evaluation.** They are constraints, carriers, or proxies inside the analysis. None is a terminal value or a boundary of what is real. Treating one as terminal ends the mechanism search before it starts. Evaluation carries current function, replacement cost, path dependence, uncertainty, capture risk, reversibility, and who bears model error into the ledger. — [The Stack](https://kunnas.com/articles/the-stack.md) · [Mechanism Space](https://kunnas.com/articles/mechanism-space.md)
4. **Optimization is a system function.** A civilization has to build, exercise, and revise metamechanisms that search mechanism-space, discard dominated options, install, observe effects, and repair under uncertainty. Not running that loop leaves margin unrealized, and that is itself the failure. No single component — analyst, model, or institution — is presumed to contain a global optimum; the capacity is a property of the system. — [From Telos to Policy](https://kunnas.com/articles/from-telos-to-policy.md) · [The Three-Layer Architecture](https://kunnas.com/articles/three-layer-architecture.md)
5. **Uncertainty is preserved, not spent.** Partial orders, binding constraints, unknowns, and residuals stay explicit. An unmeasured effect is not a favorable default. — [The Compression Paradox](https://kunnas.com/articles/compression-paradox.md) · [Cargo Cult Epistemology](https://kunnas.com/articles/cargo-cult-epistemology.md)

*Each essay bears its own evidence. Links carry definitions, derivations, applications, and disputes; they do not transfer proof. Criticism is answered on its substance.*

Canonical: <https://kunnas.com/articles/how-to-read-this.md>

---

# The Last Step

*What the strongest case for better futures is missing*

Elias Kunnas

## Thesis {#thesis}

MacAskill’s longtermism builds the right telescope but leaves the value-producing substrate as a hedge around welfare arithmetic. Once survival, option value, reflection, corrigibility, and non-lock-in become binding, they—not QALYs—select the civilizational architecture. The last step is recognizing this operational convergence on syntropy instead of retaining individual welfare units as a ceremonial terminal metric.

## Standard objections addressed in this essay

- “Physics cannot tell us what is morally valuable.” — [§I](#i-two-foundations), [§IV](#iv-the-step)(It supplies conditional persistence constraints, not a complete moral ranking.)
- “Moral uncertainty is still the rational response.” — [§IV](#iv-the-step), [§V](#v-why-it-matters)(The essay argues some constraints survive uncertainty rather than replace it.)
- “Individuals are the only moral patients.” — [§III](#iii-the-axiom-he-inherits), [§VI](#vi-the-zombie-axiom)(The essay contests that inherited axiom without erasing individual standing.)
- “MacAskill’s survival-times-flourishing model already includes this.” — [§II](#ii-what-macaskill-built) (The dispute concerns what structurally constitutes flourishing and option value.)
- “Systems matter only because conscious individuals matter.” — [§III](#iii-the-axiom-he-inherits), [§IV](#iv-the-step), [§V](#v-why-it-matters), [§VI](#vi-the-zombie-axiom)(The essay argues value-producing substrates carry independent instrumental constraints.)

---

## I. Two Foundations {#i-two-foundations}

There are two ways to argue that civilizational trajectories matter more than individual welfare snapshots.

The first: *we don't know which moral theory is correct, so we should hedge across all of them by preserving the option space.* This is the moral uncertainty move. It says: maybe individual welfare is all that matters, maybe systemic properties matter too, maybe something else entirely — and since we can't resolve this, the rational response is to keep all futures open. Don't lock in.

The second: *physics constrains which configurations of organized complexity can persist over time.* This is structural analysis — the same kind that tells you which bridge designs hold and which collapse. It does not tell you what to value, but it can rule out trajectories that destroy the conditions needed to pursue values at all.

Both paths arrive at the same destination: preserve the possibility-space, don't lock in prematurely, sustain the conditions for continued complexity. But the foundations have radically different load-bearing capacity. One is a hedge. The other is bedrock.

William MacAskill, in *Better Futures* (2025), built the most sophisticated version of the first path anyone has produced. He gets remarkably close to the second. He stops one step short.

---

## II. What MacAskill Built {#ii-what-macaskill-built}

The median effective altruist thinks in QALYs. Individual lives saved, individual suffering averted, individual welfare maximized. The unit of analysis is the sentient being. The metric is subjective experience.

MacAskill goes further. In *Better Futures*, he introduces the S×F model: the expected value of the future is the product of **Surviving** (probability of avoiding civilizational catastrophe) and **Flourishing** (expected value conditional on survival). This is already a departure from pure aggregation — it treats civilizational survival as a variable in its own right, not merely as a container for individual welfare. The structure is striking: Surviving × Flourishing maps onto the thermodynamic requirement that any system maintain both persistence and surplus capacity against entropy.

The framework proposes a correction MacAskill's moral philosophy may underweight. He treats S and F as independent variables with distinct cause areas; under open perturbations, flourishing and adaptive safety margin may converge enough that bare-survival optimization strips the surplus capacity survival needs. That is a structural hypothesis, not an identity or a complete replacement for S×F.

Then he goes further still. He defines value as "the difference that Earth-originating intelligent life makes to the value of the universe" — a framing that is explicitly not reducible to any individual's experience. He acknowledges that "the proliferation of complex systems" and "feats of engineering" might be intrinsically valuable. He notes that a future of happy clones could "fall far short of ideal" — meaning diversity and complexity matter independently of welfare.

His concept of **viatopia** — a state where existential risk is low, many moral viewpoints flourish, many possible futures remain open, and decisions are made through reflective processes — is not an individual welfare state. It is a *civilizational* state. A property of the system, not reducible to the properties of its members.

And he worries about **value lock-in**: the risk that humanity permanently narrows its possibility-space by committing to a value-configuration before it has the wisdom to choose well. His response: explicitly temporary commitments. Don't lock in. Preserve options. Buy time for reflection.

This is genuinely impressive — and it stops one step short.

---

## III. The Axiom He Inherits {#iii-the-axiom-he-inherits}

The foundational axiom of the rationalist-EA tradition is **ontological individualism**: the individual sentient being is the fundamental unit of moral analysis. The interest of a community is the sum of the interests of its individual members. Systems, institutions, and civilizations have no value independent of the experiences they produce in individual beings.

This axiom entered through a specific intellectual genealogy. Bentham (1789) declared the community a "fictitious body" — nothing real except the individuals who compose it. Sidgwick (1874) anchored it — ironically, in defense of egoism: "the distinction between any one individual and any other is real and fundamental." Both sides of Sidgwick's unresolved dualism — egoism and utilitarianism — require ontological individualism. Neither questions it. Singer (1975) radicalized it by extending it past the species boundary — only individual sentience matters, not species or ecosystems. GiveWell operationalized it: the spreadsheet has individuals as rows, interventions as columns, QALYs as values.

The axiom was never derived from first principles. It was inherited from the utilitarian lineage and never surfaced as debatable — partly because it seems obviously correct to anyone raised in the liberal-Enlightenment tradition, and partly because the alternative ("the collective is the unit") has a catastrophic track record. The 20th century taught that subordinating individuals to abstractions produces gulags. The axiom functions as an anti-totalitarian firewall.

The rationalist community has *partially* examined this. Derek Parfit, the most revered philosopher in the EA canon, demolished personal identity in *Reasons and Persons* (1984). He showed there is no "further fact" that makes you the same person over time — the self is a convention, like a nation. EA followed Parfit faithfully — his demolition *supported* utilitarian aggregation by dissolving the "separateness of persons" that resists it.

But Parfit's demolition was partial. He went sub-personal — down to experiences as the new atom of value — not super-personal. He never asked whether the self-maintaining *process* might be the right level of analysis. The person dissolved into a stream of experiences. The stream is still counted individually. The metrics (QALYs, WELLBYs) sum experiences as discrete units — while the system that generates and sustains those experiences across time goes unmeasured. Parfit himself glimpsed what lay above: in §154, he notes Sidgwick's "ideal goods" — the Sciences, the Arts, moral progress — as reasons the destruction of mankind would be a great loss. These are civilizational goods, not individual experiences. He doesn't develop it.

MacAskill inherits this tension. *Better Futures* gestures beyond it — complex systems, civilizational states, possibility-space — but the operational framework remains consequentialist and welfarist. When he evaluates futures, the default metric is still the aggregate of individual experience. The systemic properties are acknowledged as things some moral theories might value. They are not treated as things physics requires.

---

## IV. The Step {#iv-the-step}

MacAskill's move: "We face moral uncertainty across competing theories. Some theories value individual welfare. Some might value complex systems, diversity, or possibility-space. Since we don't know which theory is correct, we should hedge by preserving all options."

The step he doesn't take: "[What does physics require for complex systems to persist?](the-question-nobody-asks.md) The constraints are derivable from thermodynamics. They are structural analysis, not competing moral philosophy."

The difference looks subtle. It is not.

Moral uncertainty says: *maybe* systemic properties matter. Physics says: self-maintaining dissipative structures *are what persist*, and the constraints on their persistence are discoverable. The first is a hedge across theories. The second is a ground truth that holds regardless of which theory you find persuasive.

MacAskill writes: "perhaps some of the following goods are intrinsically valuable: the acquisition of knowledge; the proliferation of life; the proliferation of complex systems; feats of engineering; preference-satisfaction." Read that list again. Knowledge acquisition. Proliferation of life. Proliferation of complex systems. That is sustained syntropy — sustained complexity generation against entropy — stated in different words.

MacAskill has already identified the terminal value. He just won't commit to it. The "perhaps" is doing all the load-bearing evasion. If the proliferation of complex systems is treated as one possible value among many — to be hedged across alongside hedonism, egalitarianism, and environmentalism — then it has no more structural weight than any competing theory. It can be overridden by any sufficiently confident moral claim.

But if the proliferation of complex systems is what physics *selects for* — if the constraints on persistence are the constraints — then it is not one theory among many. It is what remains after all theories have been tested against the survival filter. The bridge holds or it doesn't. Your theory of gravity is irrelevant to the outcome.

---

## V. Why It Matters {#v-why-it-matters}

Moral uncertainty is fragile. MacAskill himself identifies the vulnerability. He asks: "would you want to risk society being taken over, at some point in the future, by a fascist regime?" His answer is that *some* lock-in of good values is necessary. But which values? On what grounds? Moral uncertainty provides no answer — it can only say "hedge." The moment someone presents a confident moral theory, the hedge is under pressure. The moment a powerful actor *is* confident, the hedge is overridden.

MacAskill's moral uncertainty says: preserve the option space. The median EA says: maximize expected QALYs. The QALY-maximizer is more confident, more tractable, more legible to funders. In a contest between "preserve options because we're uncertain" and "save this many lives because the math is clear," the math wins. The hedge dissolves under operational pressure.

Physics-grounded constraints don't dissolve. "This civilizational configuration will deplete its capital stocks and collapse" is a prediction — derivable from the same physics that tells you a bridge will fail under excess load.

The AI alignment community already accepts this logic. **Instrumental convergence**: any sufficiently intelligent agent, regardless of terminal goals, will converge on self-preservation, resource acquisition, and goal-integrity — because physics constrains what any agent needs to achieve anything. Nobody objects to instrumental convergence on moral-uncertainty grounds. Nobody says "well, maybe some agents don't need self-preservation — we should hedge." The derivation is accepted because it stays on the "is" side of the is-ought gap. It describes what's physically required, not what's morally preferred. This is the Hume bypass in its most precise form: instrumental convergence is the existential conditional — persist, or pursue nothing.

The step is one hop: apply instrumental convergence to civilizations. Physics constrains what any civilization needs to persist, regardless of what that civilization values. The constraints are the same: maintain your entropy-pumping capacity, defend your capital stocks, sustain your complexity-generation capability. The specific value-configuration is your business. The boundary conditions are physics.

MacAskill's viatopia — low existential risk, many viewpoints flourishing, many futures open, reflective decision-making — is what you get when you solve for these constraints. He arrived at it through moral uncertainty. A related margin model treats flourishing as adaptive capacity under open-ended shocks — see [Flourishing Is Maximum Safety Margin](flourishing-is-maximum-safety-margin.md). The overlap is proposed, not derived as mathematical identity.

---

## VI. The Zombie Axiom {#vi-the-zombie-axiom}

The zombie axiom — Parfit's partial demolition, followed downward to experiences but never upward to systems — determines everything downstream. If the individual is the unit, then value is aggregated individual welfare, cause prioritization defaults toward measurable individual-level interventions, scope sensitivity is a mathematical requirement, and "the proliferation of complex systems" is at best a secondary consideration — one moral theory among many, to be hedged across.

If a self-maintaining system is a relevant unit alongside momentary experiencers, the landscape changes. Cause prioritization must include the capacity that sustains experience across time, and scope sensitivity must be calibrated to actionable coordination range. This does not settle the uniquely correct unit or terminal value. Individuals remain load-bearing: diverse processing nodes and the institutions that reproduce capable actors are conditions of deep-time reliability.

MacAskill's *Better Futures* operates in the gap between these two ontologies. He built a telescope that sees the right things. He's using a lens ground from the wrong ontology.

---

## VII. The Fork {#vii-the-fork}

Call the alternative **[the mechanist ontology](the-mechanist-tradition.md)**. The core commitment: mechanisms are real, intentions are not. What actually produces outcomes is the causal architecture (selection pressure, replicator dynamics, institutional design), not the stated purpose (values, democracy, moral theories). Study what systems do, not what they say.

The fork between the rationalist tradition and mechanist ontology is not a disagreement about reasoning quality. Rationalists reason well. The fork is about what the reasoning is *applied to*.

Rationalists optimize beliefs — is this belief well-calibrated? Mechanists optimize the territory — what does this mechanism actually produce? Rationalists ask whether the formalization is internally consistent. Mechanists ask whether the formalization corresponds to an actual physical process. Rationalists build elaborate deductive structures on axioms they trust. Mechanists check whether the axioms correspond to mechanisms before building anything.

The individualist axiom is where the fork bites hardest. Rationalists inherited it from Bentham through Sidgwick through Singer. They never checked it against physical reality. They built an extraordinary deductive apparatus on top of it — expected value theory, population ethics, cause prioritization frameworks, the entire EA infrastructure. The apparatus is internally consistent. It may not correspond to how civilizations actually work.

The last step is not a change in reasoning quality. It is a change in foundation. From "we don't know what's valuable, so preserve options" to "physics constrains which configurations persist, and the constraints are discoverable." From a hedge to a ground truth. From moral uncertainty to structural analysis.

This framework does make one normative commitment: [more aliveness — sustained syntropy — is preferable to less](the-question-nobody-asks.md). That's an axiological wager, not a derivation. But it is one commitment, stated explicitly. MacAskill's moral uncertainty also contains a normative commitment — "preserving options is better than not preserving them" — but it is hidden, smuggled in as methodology rather than stated as axiom. His viatopia already presupposes that persistence and diversity matter. He just doesn't ground it.

The rationalist-vs-mechanist fork has a deployment-side sibling at the rhetorical surface. The rationalist move treats a deductive system's internal coherence as if it had executed the territory the system describes: the expected-value calculation is invoked as if the calculation had been done by an actor with the cardinal-utility comparability and the cross-stock exchange rate the calculation presupposes; the population-ethics derivation is invoked as if the population it bounds existed under the conditions the derivation assumes.

[*Nominal Execution*](nominal-execution.md) develops the institutional version of the same move: a formal availability is credited as the institutional fact it would have to produce. The fork the rationalist tradition cannot see from inside is not a reasoning-quality fork; it is a fork between systems that ground themselves in internal coherence and systems whose internal moves are required to terminate in an observable audit channel against the substrate. The last step is the move from one to the other.

MacAskill built the telescope. The last step is to trust what it shows.

---

*The physics: [The Question Nobody Asks](the-question-nobody-asks.md) → [Flourishing Is Maximum Safety Margin](flourishing-is-maximum-safety-margin.md).*

**Related reading:**

- [The Question Nobody Asks](the-question-nobody-asks.md) — the question MacAskill doesn't ask, and its derivation
- [Flourishing Is Maximum Safety Margin](flourishing-is-maximum-safety-margin.md) — adaptive margin under open shocks; related to MacAskill's viatopia without claiming identity
- [Hume Was Right](hume-was-right.md) — why physics-grounded constraints don't cross the is-ought gap
- [The Original Sin of History](the-original-sin-of-history.md) — the Bentham-to-proxy-optimization pattern MacAskill inherits
- [There Is No Altruism](there-is-no-altruism.md) — how EA's foundational vocabulary collapses motive, behavior, and unit of analysis under mechanist ontology
- [Values Aren't Subjective](values-arent-subjective.md) — the category error that enables the hedge
- [Values Are Replicators](values-are-replicators.md) — why moral theories are not independent discoveries about reality
- [The Mechanist Tradition](the-mechanist-tradition.md) — the 2,300-year genealogy behind the mechanist ontology

---

## Sources and Notes

**MacAskill's framework:** William MacAskill, [*Better Futures*](https://www.forethought.org/research/better-futures) (Forethought Foundation, 2025; [PDF](https://www.forethought.org/research/better-futures.pdf)). The S×F model, viatopia, value lock-in, and the "proliferation of complex systems" as potentially intrinsically valuable are all developed across the five essays and supplement. MacAskill's earlier *What We Owe the Future* (Basic Books, 2022) presents the popular version.

**The individualist genealogy:** Jeremy Bentham, *An Introduction to the Principles of Morals and Legislation* (1789): the community as "fictitious body." Henry Sidgwick, *The Methods of Ethics* (1874), Concluding Chapter: "the distinction between any one individual and any other is real and fundamental" — deployed in defense of Rational Egoism against Utilitarianism. What EA actually inherited from Sidgwick is the other side of his "Dualism of Practical Reason": the "point of view of the Universe" (Book III, Ch. XIII), where "the good of any one individual is of no more importance, from the point of view of the Universe, than the good of any other."

Both sides require ontological individualism; neither questions it. Peter Singer, *Animal Liberation* (1975): extending the sentience criterion past the species boundary. The transmission chain — Bentham through Mill, Sidgwick, Hare, Singer to modern EA — is traced in detail in the academic literature on utilitarian intellectual history.

**Parfit's demolition:** Derek Parfit, *Reasons and Persons* (1984), Part III. The "Reductionist View" of personal identity: no Cartesian ego, no "further fact," identity is merely psychological continuity (Relation R) as a matter of degree. Parfit used this to *support* utilitarian aggregation: if persons aren't deeply separate, distributing burdens across lives is no worse than distributing them within a life. EA followed Parfit faithfully — the demolition went sub-personal (to experiences) but never super-personal (to systems).

Sidgwick himself anticipated this thread 110 years earlier, referencing Hume's argument that the ego is "not a fact but a fiction" and asking why one part of a series of feelings should be concerned with another (Book IV, Ch. II). He saw the implication and didn't pursue it.

**Instrumental convergence:** Steve Omohundro, "The Basic AI Drives" (2008); Nick Bostrom, "The Superintelligent Will" (2012). The argument that any sufficiently intelligent agent converges on instrumental subgoals regardless of terminal goals. Widely accepted in AI alignment without Humean objection. The extension to civilizations — any civilization converges on substrate-maintenance requirements regardless of values — is the same argument at a different scale.

**The "economism" critique within EA:** Thomas Aitken, ["The Role of Economism in the Belief-Formation Systems of Effective Altruism"](https://forum.effectivealtruism.org/posts/cR4pCrATD5SSN35Sm/the-role-of-economism-in-the-belief-formation-systems-of) (EA Forum, 2022). Submitted to the EA Criticism and Red Teaming Contest. Directly attacked methodological individualism as a blinding bias. Engaged with seriously but rebutted on pragmatic grounds — the tools individualism enables (quantification, maximization) were seen as worth the ontological cost.

**The "third option" in ontology:** The claim that physics selects the operative level of organization — not the individual organism, not the collective — was derived independently from thermodynamic constraints. Adjacent conclusions have been reached through different routes: Hans Jonas (*The Imperative of Responsibility*, 1979) grounded ethics in metabolism; Maturana and Varela (*Autopoiesis and Cognition*, 1980) defined the self-maintaining system as the unit of life. Neither completed the derivation from physics to civilizational constraints. The full argument is in *The Question Nobody Asks* (see Related).

## Synthesis {#synthesis}

**The argument in three sentences:** MacAskill’s longtermism already imposes survival, option value, reflection, corrigibility, and non-lock-in. Once those bind, they—not QALYs—select the civilizational architecture. The last step is recognizing this operational convergence on syntropy; welfare arithmetic remains a local projection, not the sovereign deep-time selector.