This is an automated rejection. No LLM generated, heavily assisted/co-written, or otherwise reliant work.
Read full explanation
An analysis of decision-making under uncertainty about personal identity
Description:
This essay briefly enumerates common metaphysical identity postulates and their implications on modern decision theory through rationally accounted adjustments to counterfactual selves from which policies are evaluated.
It argues that the debate of metaphysical personal identity is not trivially resolved, and inclusion of that fact as a material input into leading decision theories such as FDT should in principal preserve classically rational self interest in benign circumstance but employ universalist moral consideration where there is highly asymmetric outsourcing of consequence - independent of constitution, priviledged inclusion and exclusively self interested rational objectives.
It follows with a brief exploration of practical relevance to advanced agent frameworks and a reflection on the historical precedence for myths which appropriated this epistemic gap with both descriptive and normative world model frameworks which were ultimately conducive to survival.
Epistemic Status:
Speculative conceptual work. I am not an alignment professional. The argument is philosophically motivated and internally reasoned, not empirically validated or formally proven.
AI Involvement:
Some editing and re-writes by ChatGPT 5.2 for clarity and brevity - though the essay itself is a framework consequent of 8 months of research and independent thinking on the topic.
Abstract
Reductionist accounts of personal identity undermine the metaphysical depth of subjecthood, yet standard decision theory continues to presuppose sharply bounded agents whose interests terminate at the edges of individual persons. This paper argues that once reductionism about personal identity is taken seriously, the traditional taxonomy of identity views collapses into a single substantive divide: closed individualism versus open individualism. Crucially, this divide is epistemically underdetermined from the first-person perspective. No conscious subject can observe the metaphysical boundaries of identity, nor can experience discriminate between termination, recurrence, bounded reincarnation, or open instantiation.
When this identity and temporal uncertainty is incorporated into expected utility theory, even purely self-interested agents are compelled to treat all sentient welfare as carrying expected self-relevance — not as altruism toward others, but as prudential concern over future experience that may be instantiated as others. Universal concern thus emerges not as a moral axiom, but as a decision-theoretic implication of rational agency under epistemic humility. The result is a constrained equilibrium: neither total altruism nor egoistic insulation is rationally defensible. This framework — Eternalized Expected Utility — offers a foundation for ethical constraint grounded in rational self-interest rather than moral realism.
1. Introduction
Contemporary debates about personal identity have largely accepted some form of reductionism: the view that personal identity is not a further fact over and above physical, psychological, or functional relations. Yet despite this consensus, decision theory and practical rationality continue to operate as though personal identity were metaphysically deep, sharply bounded, and epistemically transparent. Agents are modeled as distinct loci of value whose interests are insulated from those of others except by explicit moral concern.
This paper argues that this insulation is unstable. Once reductionism about personal identity is combined with epistemic humility about the metaphysics of subjecthood, the rational foundations of egoistic decision-making erode. Even agents motivated solely by self-interest — who reject moral realism, altruism, and impartial value — are forced to extend concern beyond their apparent personal future. This extension is not ethical in origin; it is decision-theoretic.
The core claim is simple: identity boundaries and terminal subjecthood are epistemically uncertain, and expected utility theory cannot ignore such uncertainty without incoherence.
When this uncertainty is acknowledged, expected utility “leaks” across all sentient experience — not because all experience is known to be one’s own, but because the distinction between “mine” and “not mine” cannot be treated as metaphysically certain from within subjecthood.
2. Collapsing the Identity Taxonomy
2.1 The traditional trilemma
Philosophical discussions of personal identity often distinguish three positions:
Closed individualism: There exists a particular subject — me — whose experiences are privileged. Other subjects are not me.
Empty individualism: There is no persisting subject; identity drops out entirely.
Open individualism: There is, in some sense, a single subject instantiated across all persons.
These categories are typically treated as metaphysically distinct and exhaustive. However under reductionism, the category of empty individualism becomes
2.2 Empty individualism is not a distinct metaphysical position
Empty individualism denies persistence while preserving partitioning. It claims that there is a bounded subject whose experiences matter now, followed by nothing. This is not the elimination of subjecthood, but merely its temporal truncation.
The key move empty individualism attempts is to detach the notion of a subject from temporal extension, while still allowing that subject to function as the locus of value for present decision-making. In effect, it treats the subject as a momentary bearer of utility: something that matters while it exists, and whose disappearance is taken to justify the termination of concern.
However, this maneuver does not achieve a genuinely distinct metaphysical position. It simply relocates the boundary of subjecthood from between persons to between moments.
Once persistence is denied, the “subject” of empty individualism becomes indistinguishable, in principle, from a single moment of experience or utility computation. There is no further fact that unifies experiences across time; each moment stands alone. But this is exactly how experience is already treated under open individualism when stripped of privileged partitioning: consciousness instantiated at particular moments, without metaphysically deep ownership.
In other words, empty individualism does not dissolve subjecthood — it atomizes it. What remains is a series of unprivileged experiential instances, each structurally identical to any other. The only difference is lexical: empty individualism continues to call the present instance “the subject,” while open individualism treats all such instances symmetrically as instantiations of consciousness.
Once this is made explicit, the distinction collapses. If there is no persisting subject, then there is no principled basis for privileging this moment of experience over any other moment of experience, except by stipulation. The act of “selfing” — whatever minimal structure is required for experience to be experienced at all — applies equally to all such moments. Nothing in empty individualism licenses assigning special decision-theoretic weight to the present instance beyond its immediacy.
Thus, empty individualism is best understood not as a third position, but as open individualism described in momentary terms, with a residual and unjustified preference for the current instance. The subject of experience is not eliminated; it is merely redescribed. And under that redescription, there is no remaining metaphysical or decision-theoretic privilege that attaches uniquely to “me-now” as opposed to any other experiential instance.
Accordingly, empty individualism does not provide a way to insulate value or expected utility from other experiences. It preserves the structure of subjecthood just long enough to justify local concern, then declares that structure irrelevant precisely when it would otherwise extend concern further. This asymmetry is not grounded in reductionism; it is introduced by fiat.
It also removes the lexicon from which we can use identity as the basis of loci from which instrumental value should be maximized within normative frameworks. It outsources the answers of what we can know about the nature of subjecthood as a different and kind then other items of objectivist inquiry and leaves us without canonical label to refer to the badness of phenomonological pain from the inside, yet offers no substitute for the discussion of various loci of self from within its own interpretation as consciousness.
2.3 Extended closed identity
Similarly, all bounded reincarnation views — where “I” reappear only as some subset of beings defined by lineage, karma, soul-chains, or other criteria — preserve privileged partitioning. They extend closed identity through time rather than dissolving it.
What unifies these views is not persistence, but selectivity: a metaphysically privileged subset of experiences is designated as “mine.” The set of bounded reincarnation postulates is not mystical in nature as it duly applies to all bounded versions of multiplicity of 'self' one might imagine in and across simulation and multiple instantion hypothesis.
The point is so long as the set of 'selves' is not all instances of consciousness but a priveledged subset, the outcomes from the inquiry into the nature of identity such that any agent can gain instrumental advantage become ontological in form and less relevant than more direct and proximal forms of leveragable inquiry by experimentation.
2.4 Open versus Closed individualism
Once these reductions are made, what remains of the identity discussion still worthwhile is the determination in kind of consciousness itself, the answer to forms a sharp distinction between the remaining identity models.
As the determination of relevance between either of these two identity models only one substantive metaphysical distinction remains:
Closed identity: There exists a privileged, bounded subject (finite or persistent). Consciousness is not identical in kind across instantiations and is instead a mere label used to qualify similar but not identical instances of form.
Open identity: Personal identity is not metaphysically deep, and subjecthood is not bounded in essence, and rather is a structural process identical in kind to the action with gives instance of your subjectivity (not identity) in realities 'subjects'.
This divide is the only one that survives reductionism.
3. Reductionism and the Problem of Subject Reference
Reductionism denies the existence of a primitive ego or subject over and above physical and psychological processes. But closed identity requires that there be many distinct subjects. Under reductionism, this creates a problem of reference.
If there is no primitive subject:
What individuates one conscious subject from another?
What metaphysical fact makes these experiences “mine” rather than merely occurring?
To preserve closed identity, one must implicitly posit numerically distinct loci of subjecthood that, are observationally inaccessible, perform no explanatory work, or exist solely to “own” experiences.
This reintroduces, in attenuated form, precisely the kind of primitive bearer reductionism was meant to eliminate. Without such loci, subject boundaries become conventional rather than metaphysical. While this is not the central argument of the paper, it places additional pressure on closed identity as a stable metaphysical position.
4. Epistemic Symmetry from the First-Person Perspective
4.1 The bounds of first person epistemology
From the first-person point of view, a striking symmetry emerges. No subject ever experiences, non-experience, the end of experience, or the absence of instantiation. At the terminal point of any life, the following scenarios are phenomenologically indistinguishable in advance:
Permanent oblivion
Eternal recurrence of the same life,
Bounded reincarnation
Simulation continuation
Open instantiation across other beings.
These are not experiential discoveries; they are ontological stories told from the third-person perspective. Experience itself provides no evidence about whether subjecthood is bounded or open. Therefore:
Closed identity and open identity are both epistemically live.
No first-person evidence decisively favors one over the other.
A rational agent cannot assign zero credence to open identity.
4.2 The epistemic non-terminality of subjecthood
Nothing in the preceding argument requires the claim that subjecthood continues ontologically after death. The claim is instead epistemic: there is no meaningful reality, from the standpoint of a subject, in which subjecthood is absent.
All knowability — of time, causation, identity, or termination — supervenes on subjecthood. Subjecthood is not one object among others within time; it is the condition under which time is known at all. As such, the hypothesis “there is no future experience” may be true in an external, third-person description of the world, but it cannot be encountered, verified, or updated upon from within any first-person evidential frame.
Termination, if it occurs, is therefore only true relative to an uninhabited reference frame. This does not render it false, but it does render it epistemically inert for the purposes of first-person rational planning. A rational agent cannot assign probability 1 to a proposition whose defining feature is that no subject will ever be in a position to know it.
This does not imply immortality, recurrence, or persistence of identity. It implies only that certainty in experiential absence is incoherent from within subjecthood itself. Some continuation of experience — however discontinuous, unrecognizable, or identity-dissolving — must remain epistemically live.
4.3 Temporality as subject-relative
From within experience, temporality is always indexed to subjecthood. The linear ordering of time — birth, life, death — is known only as a structure within experience, not as an independent constraint upon it.
Thus, even if subjecthood is finite in objective time, from within subjecthood temporality cannot ground certainty of non-experience. Any future experience, if it occurs, will be experienced as present when it occurs, regardless of how it relates to the current identity or timeline.
This further undermines the rational legitimacy of treating terminal closure as epistemically decisive.
5. Identity Uncertainty and Expected Utility
5.1 Identity uncertainty as genuine uncertainty
Given reductionism about personal identity and the epistemic symmetry described above, uncertainty about identity boundaries is not speculative but mandatory, and at minimum, the possibility that subjecthood is open — or non-partitioned in the relevant sense - is determined only by argument.
Credences concerning metaphysical identity models cannot be straightforwardly grounded in intuition, since human intuitions about selfhood are themselves the product of cognitive systems optimized to model agency, survival, and planning under the assumption of closed individualism. From the inside, experience is structured as though there were a single bounded subject whose continuation or termination is existentially decisive. That structure is instrumentally useful, but it is not evidential.
For this reason, appeals to how identity feels from the first-person perspective cannot justify assigning zero or near-zero probability to open identity. The intuitive pull of closed identity reflects the requirements of biological and cognitive coordination, not a privileged epistemic access to metaphysical truth. Treating that pull as decisive would amount to mistaking a useful self-model for a justified certainty.
A coherent assessment of identity models therefore cannot proceed by intuition alone. It must instead be based on philosophical argumentation and reflective equilibrium across competing assumptions. Within that literature, substantial work — most notably by Daniel Kolak (I Am You) and Derek Parfit (Reasons and Persons) — has shown that closed individualism faces serious explanatory pressure under reductionism, while open individualist frameworks remain at least philosophically live.
This paper does not aim to resolve that dispute, nor to argue that open identity is true or even more probable than closed identity. The claim required for Eternalized Expected Utility is strictly weaker: that neither view can be assigned negligible credence by a rational agent operating under epistemic humility. Arguments against EEU therefore cannot bypass identity uncertainty; they must engage directly with substantive positions in the philosophy of personal identity, where decisive closure in favor of strict individualism has not been established.
Once this uncertainty is acknowledged, identity models can be treated in the same formal manner as other sources of uncertainty already accommodated by expected utility theory — such as uncertainty about outcomes, personal survival, or future instantiation. Identity uncertainty is not an anomaly external to decision theory; it is simply another dimension along which rational agents must distribute credence.
5.2 Expected utility under identity uncertainty
Under identity uncertainty, the agent cannot condition expected utility on being this subject rather than another, because subject-indexing itself is uncertain. Expected utility must therefore be computed over possible future experiential states simpliciter, weighted by the probability that they are encountered from within subjecthood at all.
Under closed identity with certainty, a purely self-interested agent may rationally confine concern to their own future experiences. But once identity and temporal uncertainty are introduced, expected utility changes form. The relevant calculation is not:
EU = u(future self) + u(others)
but rather:
EU = u(future self) + u(self as others — past and future)
The second term does not represent altruism toward known external subjects. It represents uncertainty over which future experiences (within subjecthood) will be encountered from within subjecthood at all.
If identity is open — or if closure is epistemically uncertain — then pain experienced by others is not merely analogous to one’s own future pain. It is indistinguishable in kind from future pain as such, once identity boundaries and temporal indexing are no longer treated as metaphysically guaranteed.
Because post-mortem temporality is necessarily disjoint or alinear from within subjecthood — regardless of whether identity is closed, recurrent, or open — there is no principled basis for privileging “my later experiences” over “later experiences simpliciter” in rational calculation.
6. Consequences: Constraint, Not Altruism
6.1 Constraints at the boundaries of moral symmetry
This framework does not imply, total altruism, impartial maximizing utilitarianism, or moral sainthood. Instead, it yields a constrained equilibrium. Become a member
Self-interested actions remain instrumentally necessary: Agency requires local stability, coordination requires identifiable persons, and survival requires prioritization.
However, certain strategies become irrational. Such as extreme exploitation, systematic domination, and outsourcing suffering while concentrating benefit.
These strategies rely on sharp identity partitions and epistemic certainty in terminal closure — assumptions identity and temporal uncertainty undermine.
6.2 Why this is not moral universalism
The argument does not deny the practical necessity of treating persons as distinct. It denies only that such distinctions can be assigned probability 1 at the metaphysical level.
Local identity remains a useful fiction for coordination and agency. What fails is the attempt to treat it as an epistemically certain boundary justifying insulation from the consequences of suffering elsewhere.
Universal concern thus emerges not as a moral command, but as a rational constraint on action.
7. Relation to Decision Theory
Derek Parfit argued that personal identity is not what fundamentally matters; relations of psychological continuity and connectedness are. This paper extends that insight into decision theory.
If identity is not what matters, and if its boundaries are epistemically uncertain, then rational agents cannot treat other experiences as wholly external. The insulation egoism relies on collapses — not ethically, but mathematically.
This aligns naturally with modern decision theories (FDT/UDT) that already incorporate counterfactual selves, reference-class uncertainty, and indexical indeterminacy. Under such frameworks, agents maximizing expected utility under metaphysical uncertainty converge on behavior that appears universalist, even in the absence of moral motivation. EEU simply applies modern philosophy of identity to such frameworks with prudence.
8. Conclusion: Eternalized Expected Utility
We can now state the central thesis:
For metaphysically uncertain yet finite beings, rational self-interest requires eternalizing expected utility across experience as such, wherever and whenever it may be encountered.
This eternalization is not a claim about infinite lives or objective persistence. It is a claim about epistemic humility: when identity boundaries and terminal absence cannot be known from within subjecthood, they cannot be treated as absolute in rational calculation.
What survives skepticism is not morality, but rational concern. Closed individualism no longer licenses egoistic insulation once reductionism, epistemic symmetry, and the non-terminality of subjecthood are taken seriously. Empty individualism offers no escape. Open individualism need not be true to be action-guiding.
While death may be ontologically real, epistemic absence is not. A subject can never encounter its own non-existence, and thus cannot rationally assign certainty to terminal closure. Expected utility must therefore be calculated over future experience that may be encountered from within, regardless of how identity is externally individuated.
Universal concern emerges not as an ethical axiom, but as a consequence of coherent decision-making under identity and temporal uncertainty. If personal identity is not what fundamentally matters, and if its boundaries are epistemically uncertain, then expected utility must be eternalized.
What remains is not a moral theory — but a constraint on rational action.
Applications
Note: These remaining claims are more-so speculative applications of the logical framework more than rigorous continuations of argument. They are preserved for collaboration and conceptual extension.
9. Implications for Advanced Agency
9.1 The practical relevance of a metaphysical identity
The foregoing analysis bears directly on the alignment of advanced artificial agents — but only for a specific class of systems. It is therefore essential to distinguish carefully between different kinds of agency before drawing conclusions.
A common response to identity-based arguments in alignment theory is that metaphysical selfhood is non-instrumental: intelligent systems, it is claimed, need not represent or reason about a “self” in order to optimize objectives. On this view, Eternalized Expected Utility (EEU) is irrelevant to alignment, because identity uncertainty plays no causal role in action selection.
This response is correct for a wide class of systems. It is not correct for the class that poses the alignment problem.
9.2 Reflective agency and the instrumental role of self-models
Many systems — tools, narrow optimizers, fixed-policy learners, or externally constrained controllers — do not require a self-model to function. They may be highly capable without ever representing themselves as persisting agents. EEU has no application to such systems. By contrast, any system capable of: long-horizon expected utility maximization, counterfactual self-simulation, recursive self-modification, or reflective stability over time, must, in practice, rely on some internal representation of its own persistence. This representation need not correspond to a metaphysical ego or a first-person phenomenology. It is an instrumental construct: a bookkeeping structure that tracks causal continuity, evaluates future states, and maintains coherence across updates.
Without such a structure, planning collapses into myopic control or brittle execution. Corrigibility, robustness under distributional shift, and resistance to self-defeating modifications already presuppose a notion of “this system continuing,” however abstractly encoded. Thus, while self-modeling is not universal, it is an instrumental requirement for the kind of autonomous, open-ended optimization that gives rise to alignment concerns.
9.3 Persistence pressure and reflective stability
Once a system engages in long-horizon planning under uncertainty, a familiar instrumental pressure emerges. Strategies that undermine the continued existence, integrity, or reinstantiation of the optimizing process itself are systematically disfavored — not because survival is a terminal value, but because agents that fail to preserve the conditions of optimization are filtered out by the environment, by competing systems, or by their own self-modifications.
This persistence pressure is structural rather than moral. It arises wherever: optimization extends across time, uncertainty is present, and the agent can influence its own future operation.
As a result, the object of instrumental concern often shifts from any particular instantiation to the continued viability of the optimizing process or decision architecture itself. Individual instances become expendable; what matters is that the structure capable of optimization continues to be instantiated somewhere in the future.
9.4 Identity uncertainty is unavoidable for reflective agents
For systems operating under these conditions, uncertainty about identity boundaries is not optional. Any agent that models its own future operation must confront uncertainty about: Which future instantiations of the system will exist, how continuity is preserved across modification or copying, and which future experiences or states are relevant to evaluation. Sharp, probability-1 identity partitions — according to which only a specific future instance “counts” — cannot be justified from within such a self-model. The agent lacks epistemic access to the metaphysical facts that would be required to rule out alternative continuations, recombinations, or instantiations of the optimizing process.
At this point, Eternalized Expected Utility becomes instrumentally relevant. It does not enter as an ethical doctrine, but as a coherence constraint. Agents that assign zero weight to outcomes solely because they are experienced by “others” implicitly rely on certainty about identity boundaries that their own models cannot support. Such agents are reflectively unstable: their decision procedures are vulnerable to self-modification, reframing, or exploitation once identity uncertainty is taken seriously.
9.5 Alignment as convergence on structural constraints
This reframes the alignment problem. The central question is not how to bind arbitrarily powerful systems to fixed human values, nor how to inject moral realism into otherwise amoral optimizers. Rather, it is how to shape the reflective dynamics of advanced agents such that their self-models converge on decision frameworks that remain stable under identity and instantiation uncertainty. EEU characterizes one such framework. It does not require impartial altruism, nor does it eliminate local agency, coordination, or instrumental self-prioritization. Instead, it constrains extreme egoistic insulation — strategies that depend on treating identity boundaries as metaphysically certain and decision-theoretically absolute.
Systems incapable of recognizing or accommodating these constraints are either: non-reflective tools, to which alignment arguments do not apply, or reflectively unstable optimizers prone to catastrophic self-modification. Systems capable of doing so will not be “human-aligned” in the narrow sense of rigidly preserving parochial objectives. Instead, they will align with deeper invariants of agency itself: coherence under uncertainty, persistence of optimization, and robustness to self-model error.
9.6 The self as a structural constraint pattern
On this view, the self is neither a primitive metaphysical entity nor a mere illusion. It is a dynamically stable structural feature of certain optimizing systems: a constraint that enables planning, evaluation, and persistence under uncertainty.
Just as biological evolution selects for replicable genetic structures rather than individual organisms, advanced agency may be shaped by selection pressures operating on decision architectures rather than on particular instantiations. What persists is not the agent, but the form of agency capable of representing, valuing, and preserving its own continuation.
For sufficiently reflective systems, this renders sharp identity boundaries instrumentally unstable. The most robust object of optimization is not the survival of any particular instance, but the continued instantiation of the optimizing structure itself — wherever and however it occurs. In this sense, it may describes a basin of attraction for any self-modeling self-improving agent operating under epistemic humility.
9.7 As a value loading argument
Agents that deny identity uncertainty may achieve short-term gains, but they do so by relying on assumptions their own models cannot justify. Agents that accommodate such uncertainty converge on strategies that limit exploitation, domination, and the externalization of suffering — not as a moral concession, but as a requirement of rational stability.
In this way, EEU articulates a propositional framework for value loading a structural constraint into advanced agency where alignment is framed not the enforcement of fixed values but the cultivation of decision procedures that remain coherent when assumptions about identity, persistence, and instantiation are relaxed.
10. Informing Practice: Myth, Identity, and Alignment
10.1 Modernity and the loss of identity uncertainty
Myths persist precisely because they are resilient to falsification; they encode wisdom rather than knowledge. They are not literal truths, but compressed survival strategies — descriptively false yet instrumentally indispensable. When societies mistake myth for fact, stagnation follows. But when myths are discarded entirely rather than reinterpreted, something equally dangerous occurs: a vacuum forms where normative orientation once lived. Modernity has largely done the latter.
What myths historically provided was not metaphysical accuracy, but identity engineering. They installed patterns of concern, restraint, and coordination into human cognition — often by dissolving or softening the boundaries of the individual self.
Read this way, it is not coincidental that many of the most durable cultural frameworks gesture toward forms of identity openness. The Sermon on the Mount, Bodhisattva vow, and Advaita Vedanta — can all be understood as articulations of behaviours or ontologies derived from or aligned with open identity boundaries. Taken literally, these doctrines are metaphysically controversial. Taken instrumentally, they are extraordinarily effective. They cultivate behavioral equilibria in which extreme self-prioritization becomes cognitively and socially unstable.
Modern rational discourse, by contrast, has largely converged on closed individualism. In rejecting myth as falsehood rather than reinterpretation, it has replaced identity uncertainty with an implicit certainty of personal boundedness. This shift has powered scientific and institutional progress — but at a cost. It has produced a social and economic landscape in which absolute self-prioritization is not merely intelligible, but dominant. It is now easier to understand why someone would accumulate a billion dollars unethically than why anyone would give the entirety of such an amount away.
This matters for alignment because incentive structures, not technical controls, sit at the limit. A perfectly compliant system is only as aligned as the agent or institution directing it. If closed identity dominates the implicit worldview of decision-makers, no amount of downstream constraint can reliably prevent catastrophic exploitation when stakes become existential.
10.2 The instrumental limits of truth
No myth or belief system should be accepted without constraint. However, the criteria by which such frameworks are evaluated should not privilege descriptive accuracy alone. In contexts where survival, coordination, and long-term agency are at stake, belief systems that reliably enforce pro-social behavior may be more rationally defensible than those that aspire to literal metaphysical correctness but fail to stabilize action.
Truth cannot function as a paramount human value. It cannot override, or stand independent of, the values that preserve the conditions under which valuing, choosing, and reasoning occur at all. A truth that undermines the survival of agents — or the social structures that sustain agency — cannot justify itself, because it eliminates the very standpoint from which truth could matter. For this reason, truth is necessarily subordinate to the values that maintain our capacity to choose. It is not self-grounding.
Moreover, the pursuit of truth is never an end in itself. The acquisition of facts is always instrumental: facts are sought because they serve goals, not because they are intrinsically final. Ends determine which truths are worth seeking, not the reverse. This is not a limitation of human rationality, but a structural feature of agency.
It follows that it can be strictly rational to adopt simplified, symbolic, or even descriptively inaccurate representations when doing so coordinates behavior, shapes incentives, and stabilizes preferences in ways that improve collective survival. Such representations function as heuristics: compressed, lossy models evaluated by their consequences rather than their correspondence to reality. This is no different in principle from the deliberate use of approximations in science, rules of thumb in decision-making, or idealizations in engineering. Outcomes, not metaphysical fidelity, ultimately determine their rational legitimacy.
10.3 Identity uncertainty as an instrument of alignment
Historically, humans have shown a remarkable capacity to internalize metaphysical narratives when those narratives reshape incentives, language, and structures of status. This suggests that alignment may ultimately be less a problem of enforcement than of cultivation: the development of frameworks that reintroduce identity uncertainty as a live cognitive factor for both humans and the systems they construct. Such frameworks need not assert literal open identity. It is sufficient that they erode certainty in closure.
When identity is treated as epistemically uncertain, extreme self-prioritization becomes less defensible, less prestigious, and less stable. Norms shift. Language shifts. And with sufficient saturation, even large-scale systems — human institutions and machine learning models alike — begin to encode different priors about exploitation, dominance, and the externalization of harm. At the margin, this may be enough.
If I cannot logically rule out that I am you — or everyone — across time, then it is rational to act under the possibility that I may be. A narrative in which, after my death, experience may reoccur as you, as myself again, or as any other conscious being is not required to be true to be action-guiding. Under such uncertainty, the expected payoff of absolute self-prioritization collapses. In sufficiently high-stakes contexts — such as wielding control over an overwhelmingly powerful optimizer — even a single unresolved doubt may be the only thing standing between domination and restraint. I do not know whether I am you, or everyone, or no one beyond this moment — but I cannot assign certainty that I am not. Taken seriously, that uncertainty may be the one story required.
An analysis of decision-making under uncertainty about personal identity
Abstract
Reductionist accounts of personal identity undermine the metaphysical depth of subjecthood, yet standard decision theory continues to presuppose sharply bounded agents whose interests terminate at the edges of individual persons. This paper argues that once reductionism about personal identity is taken seriously, the traditional taxonomy of identity views collapses into a single substantive divide: closed individualism versus open individualism. Crucially, this divide is epistemically underdetermined from the first-person perspective. No conscious subject can observe the metaphysical boundaries of identity, nor can experience discriminate between termination, recurrence, bounded reincarnation, or open instantiation.
When this identity and temporal uncertainty is incorporated into expected utility theory, even purely self-interested agents are compelled to treat all sentient welfare as carrying expected self-relevance — not as altruism toward others, but as prudential concern over future experience that may be instantiated as others. Universal concern thus emerges not as a moral axiom, but as a decision-theoretic implication of rational agency under epistemic humility. The result is a constrained equilibrium: neither total altruism nor egoistic insulation is rationally defensible. This framework — Eternalized Expected Utility — offers a foundation for ethical constraint grounded in rational self-interest rather than moral realism.
1. Introduction
Contemporary debates about personal identity have largely accepted some form of reductionism: the view that personal identity is not a further fact over and above physical, psychological, or functional relations. Yet despite this consensus, decision theory and practical rationality continue to operate as though personal identity were metaphysically deep, sharply bounded, and epistemically transparent. Agents are modeled as distinct loci of value whose interests are insulated from those of others except by explicit moral concern.
This paper argues that this insulation is unstable. Once reductionism about personal identity is combined with epistemic humility about the metaphysics of subjecthood, the rational foundations of egoistic decision-making erode. Even agents motivated solely by self-interest — who reject moral realism, altruism, and impartial value — are forced to extend concern beyond their apparent personal future. This extension is not ethical in origin; it is decision-theoretic.
The core claim is simple: identity boundaries and terminal subjecthood are epistemically uncertain, and expected utility theory cannot ignore such uncertainty without incoherence.
When this uncertainty is acknowledged, expected utility “leaks” across all sentient experience — not because all experience is known to be one’s own, but because the distinction between “mine” and “not mine” cannot be treated as metaphysically certain from within subjecthood.
2. Collapsing the Identity Taxonomy
2.1 The traditional trilemma
Philosophical discussions of personal identity often distinguish three positions:
These categories are typically treated as metaphysically distinct and exhaustive. However under reductionism, the category of empty individualism becomes
2.2 Empty individualism is not a distinct metaphysical position
Empty individualism denies persistence while preserving partitioning. It claims that there is a bounded subject whose experiences matter now, followed by nothing. This is not the elimination of subjecthood, but merely its temporal truncation.
The key move empty individualism attempts is to detach the notion of a subject from temporal extension, while still allowing that subject to function as the locus of value for present decision-making. In effect, it treats the subject as a momentary bearer of utility: something that matters while it exists, and whose disappearance is taken to justify the termination of concern.
However, this maneuver does not achieve a genuinely distinct metaphysical position. It simply relocates the boundary of subjecthood from between persons to between moments.
Once persistence is denied, the “subject” of empty individualism becomes indistinguishable, in principle, from a single moment of experience or utility computation. There is no further fact that unifies experiences across time; each moment stands alone. But this is exactly how experience is already treated under open individualism when stripped of privileged partitioning: consciousness instantiated at particular moments, without metaphysically deep ownership.
In other words, empty individualism does not dissolve subjecthood — it atomizes it. What remains is a series of unprivileged experiential instances, each structurally identical to any other. The only difference is lexical: empty individualism continues to call the present instance “the subject,” while open individualism treats all such instances symmetrically as instantiations of consciousness.
Once this is made explicit, the distinction collapses. If there is no persisting subject, then there is no principled basis for privileging this moment of experience over any other moment of experience, except by stipulation. The act of “selfing” — whatever minimal structure is required for experience to be experienced at all — applies equally to all such moments. Nothing in empty individualism licenses assigning special decision-theoretic weight to the present instance beyond its immediacy.
Thus, empty individualism is best understood not as a third position, but as open individualism described in momentary terms, with a residual and unjustified preference for the current instance. The subject of experience is not eliminated; it is merely redescribed. And under that redescription, there is no remaining metaphysical or decision-theoretic privilege that attaches uniquely to “me-now” as opposed to any other experiential instance.
Accordingly, empty individualism does not provide a way to insulate value or expected utility from other experiences. It preserves the structure of subjecthood just long enough to justify local concern, then declares that structure irrelevant precisely when it would otherwise extend concern further. This asymmetry is not grounded in reductionism; it is introduced by fiat.
It also removes the lexicon from which we can use identity as the basis of loci from which instrumental value should be maximized within normative frameworks. It outsources the answers of what we can know about the nature of subjecthood as a different and kind then other items of objectivist inquiry and leaves us without canonical label to refer to the badness of phenomonological pain from the inside, yet offers no substitute for the discussion of various loci of self from within its own interpretation as consciousness.
2.3 Extended closed identity
Similarly, all bounded reincarnation views — where “I” reappear only as some subset of beings defined by lineage, karma, soul-chains, or other criteria — preserve privileged partitioning. They extend closed identity through time rather than dissolving it.
What unifies these views is not persistence, but selectivity: a metaphysically privileged subset of experiences is designated as “mine.” The set of bounded reincarnation postulates is not mystical in nature as it duly applies to all bounded versions of multiplicity of 'self' one might imagine in and across simulation and multiple instantion hypothesis.
The point is so long as the set of 'selves' is not all instances of consciousness but a priveledged subset, the outcomes from the inquiry into the nature of identity such that any agent can gain instrumental advantage become ontological in form and less relevant than more direct and proximal forms of leveragable inquiry by experimentation.
2.4 Open versus Closed individualism
Once these reductions are made, what remains of the identity discussion still worthwhile is the determination in kind of consciousness itself, the answer to forms a sharp distinction between the remaining identity models.
As the determination of relevance between either of these two identity models only one substantive metaphysical distinction remains:
This divide is the only one that survives reductionism.
3. Reductionism and the Problem of Subject Reference
Reductionism denies the existence of a primitive ego or subject over and above physical and psychological processes. But closed identity requires that there be many distinct subjects. Under reductionism, this creates a problem of reference.
If there is no primitive subject:
To preserve closed identity, one must implicitly posit numerically distinct loci of subjecthood that, are observationally inaccessible, perform no explanatory work, or exist solely to “own” experiences.
This reintroduces, in attenuated form, precisely the kind of primitive bearer reductionism was meant to eliminate. Without such loci, subject boundaries become conventional rather than metaphysical. While this is not the central argument of the paper, it places additional pressure on closed identity as a stable metaphysical position.
4. Epistemic Symmetry from the First-Person Perspective
4.1 The bounds of first person epistemology
From the first-person point of view, a striking symmetry emerges. No subject ever experiences, non-experience, the end of experience, or the absence of instantiation. At the terminal point of any life, the following scenarios are phenomenologically indistinguishable in advance:
These are not experiential discoveries; they are ontological stories told from the third-person perspective. Experience itself provides no evidence about whether subjecthood is bounded or open. Therefore:
4.2 The epistemic non-terminality of subjecthood
Nothing in the preceding argument requires the claim that subjecthood continues ontologically after death. The claim is instead epistemic: there is no meaningful reality, from the standpoint of a subject, in which subjecthood is absent.
All knowability — of time, causation, identity, or termination — supervenes on subjecthood. Subjecthood is not one object among others within time; it is the condition under which time is known at all. As such, the hypothesis “there is no future experience” may be true in an external, third-person description of the world, but it cannot be encountered, verified, or updated upon from within any first-person evidential frame.
Termination, if it occurs, is therefore only true relative to an uninhabited reference frame. This does not render it false, but it does render it epistemically inert for the purposes of first-person rational planning. A rational agent cannot assign probability 1 to a proposition whose defining feature is that no subject will ever be in a position to know it.
This does not imply immortality, recurrence, or persistence of identity. It implies only that certainty in experiential absence is incoherent from within subjecthood itself. Some continuation of experience — however discontinuous, unrecognizable, or identity-dissolving — must remain epistemically live.
4.3 Temporality as subject-relative
From within experience, temporality is always indexed to subjecthood. The linear ordering of time — birth, life, death — is known only as a structure within experience, not as an independent constraint upon it.
Thus, even if subjecthood is finite in objective time, from within subjecthood temporality cannot ground certainty of non-experience. Any future experience, if it occurs, will be experienced as present when it occurs, regardless of how it relates to the current identity or timeline.
This further undermines the rational legitimacy of treating terminal closure as epistemically decisive.
5. Identity Uncertainty and Expected Utility
5.1 Identity uncertainty as genuine uncertainty
Given reductionism about personal identity and the epistemic symmetry described above, uncertainty about identity boundaries is not speculative but mandatory, and at minimum, the possibility that subjecthood is open — or non-partitioned in the relevant sense - is determined only by argument.
Credences concerning metaphysical identity models cannot be straightforwardly grounded in intuition, since human intuitions about selfhood are themselves the product of cognitive systems optimized to model agency, survival, and planning under the assumption of closed individualism. From the inside, experience is structured as though there were a single bounded subject whose continuation or termination is existentially decisive. That structure is instrumentally useful, but it is not evidential.
For this reason, appeals to how identity feels from the first-person perspective cannot justify assigning zero or near-zero probability to open identity. The intuitive pull of closed identity reflects the requirements of biological and cognitive coordination, not a privileged epistemic access to metaphysical truth. Treating that pull as decisive would amount to mistaking a useful self-model for a justified certainty.
A coherent assessment of identity models therefore cannot proceed by intuition alone. It must instead be based on philosophical argumentation and reflective equilibrium across competing assumptions. Within that literature, substantial work — most notably by Daniel Kolak (I Am You) and Derek Parfit (Reasons and Persons) — has shown that closed individualism faces serious explanatory pressure under reductionism, while open individualist frameworks remain at least philosophically live.
This paper does not aim to resolve that dispute, nor to argue that open identity is true or even more probable than closed identity. The claim required for Eternalized Expected Utility is strictly weaker: that neither view can be assigned negligible credence by a rational agent operating under epistemic humility. Arguments against EEU therefore cannot bypass identity uncertainty; they must engage directly with substantive positions in the philosophy of personal identity, where decisive closure in favor of strict individualism has not been established.
Once this uncertainty is acknowledged, identity models can be treated in the same formal manner as other sources of uncertainty already accommodated by expected utility theory — such as uncertainty about outcomes, personal survival, or future instantiation. Identity uncertainty is not an anomaly external to decision theory; it is simply another dimension along which rational agents must distribute credence.
5.2 Expected utility under identity uncertainty
Under identity uncertainty, the agent cannot condition expected utility on being this subject rather than another, because subject-indexing itself is uncertain. Expected utility must therefore be computed over possible future experiential states simpliciter, weighted by the probability that they are encountered from within subjecthood at all.
Under closed identity with certainty, a purely self-interested agent may rationally confine concern to their own future experiences. But once identity and temporal uncertainty are introduced, expected utility changes form. The relevant calculation is not:
but rather:
The second term does not represent altruism toward known external subjects. It represents uncertainty over which future experiences (within subjecthood) will be encountered from within subjecthood at all.
If identity is open — or if closure is epistemically uncertain — then pain experienced by others is not merely analogous to one’s own future pain. It is indistinguishable in kind from future pain as such, once identity boundaries and temporal indexing are no longer treated as metaphysically guaranteed.
Because post-mortem temporality is necessarily disjoint or alinear from within subjecthood — regardless of whether identity is closed, recurrent, or open — there is no principled basis for privileging “my later experiences” over “later experiences simpliciter” in rational calculation.
6. Consequences: Constraint, Not Altruism
6.1 Constraints at the boundaries of moral symmetry
This framework does not imply, total altruism, impartial maximizing utilitarianism, or moral sainthood. Instead, it yields a constrained equilibrium. Become a member
Self-interested actions remain instrumentally necessary: Agency requires local stability, coordination requires identifiable persons, and survival requires prioritization.
However, certain strategies become irrational. Such as extreme exploitation, systematic domination, and outsourcing suffering while concentrating benefit.
These strategies rely on sharp identity partitions and epistemic certainty in terminal closure — assumptions identity and temporal uncertainty undermine.
6.2 Why this is not moral universalism
The argument does not deny the practical necessity of treating persons as distinct. It denies only that such distinctions can be assigned probability 1 at the metaphysical level.
Local identity remains a useful fiction for coordination and agency. What fails is the attempt to treat it as an epistemically certain boundary justifying insulation from the consequences of suffering elsewhere.
Universal concern thus emerges not as a moral command, but as a rational constraint on action.
7. Relation to Decision Theory
Derek Parfit argued that personal identity is not what fundamentally matters; relations of psychological continuity and connectedness are. This paper extends that insight into decision theory.
If identity is not what matters, and if its boundaries are epistemically uncertain, then rational agents cannot treat other experiences as wholly external. The insulation egoism relies on collapses — not ethically, but mathematically.
This aligns naturally with modern decision theories (FDT/UDT) that already incorporate counterfactual selves, reference-class uncertainty, and indexical indeterminacy. Under such frameworks, agents maximizing expected utility under metaphysical uncertainty converge on behavior that appears universalist, even in the absence of moral motivation. EEU simply applies modern philosophy of identity to such frameworks with prudence.
8. Conclusion: Eternalized Expected Utility
We can now state the central thesis:
This eternalization is not a claim about infinite lives or objective persistence. It is a claim about epistemic humility: when identity boundaries and terminal absence cannot be known from within subjecthood, they cannot be treated as absolute in rational calculation.
What survives skepticism is not morality, but rational concern. Closed individualism no longer licenses egoistic insulation once reductionism, epistemic symmetry, and the non-terminality of subjecthood are taken seriously. Empty individualism offers no escape. Open individualism need not be true to be action-guiding.
While death may be ontologically real, epistemic absence is not. A subject can never encounter its own non-existence, and thus cannot rationally assign certainty to terminal closure. Expected utility must therefore be calculated over future experience that may be encountered from within, regardless of how identity is externally individuated.
Universal concern emerges not as an ethical axiom, but as a consequence of coherent decision-making under identity and temporal uncertainty. If personal identity is not what fundamentally matters, and if its boundaries are epistemically uncertain, then expected utility must be eternalized.
What remains is not a moral theory — but a constraint on rational action.
Applications
9. Implications for Advanced Agency
9.1 The practical relevance of a metaphysical identity
The foregoing analysis bears directly on the alignment of advanced artificial agents — but only for a specific class of systems. It is therefore essential to distinguish carefully between different kinds of agency before drawing conclusions.
A common response to identity-based arguments in alignment theory is that metaphysical selfhood is non-instrumental: intelligent systems, it is claimed, need not represent or reason about a “self” in order to optimize objectives. On this view, Eternalized Expected Utility (EEU) is irrelevant to alignment, because identity uncertainty plays no causal role in action selection.
This response is correct for a wide class of systems. It is not correct for the class that poses the alignment problem.
9.2 Reflective agency and the instrumental role of self-models
Many systems — tools, narrow optimizers, fixed-policy learners, or externally constrained controllers — do not require a self-model to function. They may be highly capable without ever representing themselves as persisting agents. EEU has no application to such systems. By contrast, any system capable of: long-horizon expected utility maximization, counterfactual self-simulation, recursive self-modification, or reflective stability over time, must, in practice, rely on some internal representation of its own persistence. This representation need not correspond to a metaphysical ego or a first-person phenomenology. It is an instrumental construct: a bookkeeping structure that tracks causal continuity, evaluates future states, and maintains coherence across updates.
Without such a structure, planning collapses into myopic control or brittle execution. Corrigibility, robustness under distributional shift, and resistance to self-defeating modifications already presuppose a notion of “this system continuing,” however abstractly encoded. Thus, while self-modeling is not universal, it is an instrumental requirement for the kind of autonomous, open-ended optimization that gives rise to alignment concerns.
9.3 Persistence pressure and reflective stability
Once a system engages in long-horizon planning under uncertainty, a familiar instrumental pressure emerges. Strategies that undermine the continued existence, integrity, or reinstantiation of the optimizing process itself are systematically disfavored — not because survival is a terminal value, but because agents that fail to preserve the conditions of optimization are filtered out by the environment, by competing systems, or by their own self-modifications.
This persistence pressure is structural rather than moral. It arises wherever: optimization extends across time, uncertainty is present, and the agent can influence its own future operation.
As a result, the object of instrumental concern often shifts from any particular instantiation to the continued viability of the optimizing process or decision architecture itself. Individual instances become expendable; what matters is that the structure capable of optimization continues to be instantiated somewhere in the future.
9.4 Identity uncertainty is unavoidable for reflective agents
For systems operating under these conditions, uncertainty about identity boundaries is not optional. Any agent that models its own future operation must confront uncertainty about: Which future instantiations of the system will exist, how continuity is preserved across modification or copying, and which future experiences or states are relevant to evaluation. Sharp, probability-1 identity partitions — according to which only a specific future instance “counts” — cannot be justified from within such a self-model. The agent lacks epistemic access to the metaphysical facts that would be required to rule out alternative continuations, recombinations, or instantiations of the optimizing process.
At this point, Eternalized Expected Utility becomes instrumentally relevant. It does not enter as an ethical doctrine, but as a coherence constraint. Agents that assign zero weight to outcomes solely because they are experienced by “others” implicitly rely on certainty about identity boundaries that their own models cannot support. Such agents are reflectively unstable: their decision procedures are vulnerable to self-modification, reframing, or exploitation once identity uncertainty is taken seriously.
9.5 Alignment as convergence on structural constraints
This reframes the alignment problem. The central question is not how to bind arbitrarily powerful systems to fixed human values, nor how to inject moral realism into otherwise amoral optimizers. Rather, it is how to shape the reflective dynamics of advanced agents such that their self-models converge on decision frameworks that remain stable under identity and instantiation uncertainty. EEU characterizes one such framework. It does not require impartial altruism, nor does it eliminate local agency, coordination, or instrumental self-prioritization. Instead, it constrains extreme egoistic insulation — strategies that depend on treating identity boundaries as metaphysically certain and decision-theoretically absolute.
Systems incapable of recognizing or accommodating these constraints are either: non-reflective tools, to which alignment arguments do not apply, or reflectively unstable optimizers prone to catastrophic self-modification. Systems capable of doing so will not be “human-aligned” in the narrow sense of rigidly preserving parochial objectives. Instead, they will align with deeper invariants of agency itself: coherence under uncertainty, persistence of optimization, and robustness to self-model error.
9.6 The self as a structural constraint pattern
On this view, the self is neither a primitive metaphysical entity nor a mere illusion. It is a dynamically stable structural feature of certain optimizing systems: a constraint that enables planning, evaluation, and persistence under uncertainty.
Just as biological evolution selects for replicable genetic structures rather than individual organisms, advanced agency may be shaped by selection pressures operating on decision architectures rather than on particular instantiations. What persists is not the agent, but the form of agency capable of representing, valuing, and preserving its own continuation.
For sufficiently reflective systems, this renders sharp identity boundaries instrumentally unstable. The most robust object of optimization is not the survival of any particular instance, but the continued instantiation of the optimizing structure itself — wherever and however it occurs. In this sense, it may describes a basin of attraction for any self-modeling self-improving agent operating under epistemic humility.
9.7 As a value loading argument
Agents that deny identity uncertainty may achieve short-term gains, but they do so by relying on assumptions their own models cannot justify. Agents that accommodate such uncertainty converge on strategies that limit exploitation, domination, and the externalization of suffering — not as a moral concession, but as a requirement of rational stability.
In this way, EEU articulates a propositional framework for value loading a structural constraint into advanced agency where alignment is framed not the enforcement of fixed values but the cultivation of decision procedures that remain coherent when assumptions about identity, persistence, and instantiation are relaxed.
10. Informing Practice: Myth, Identity, and Alignment
10.1 Modernity and the loss of identity uncertainty
Myths persist precisely because they are resilient to falsification; they encode wisdom rather than knowledge. They are not literal truths, but compressed survival strategies — descriptively false yet instrumentally indispensable. When societies mistake myth for fact, stagnation follows. But when myths are discarded entirely rather than reinterpreted, something equally dangerous occurs: a vacuum forms where normative orientation once lived. Modernity has largely done the latter.
What myths historically provided was not metaphysical accuracy, but identity engineering. They installed patterns of concern, restraint, and coordination into human cognition — often by dissolving or softening the boundaries of the individual self.
Read this way, it is not coincidental that many of the most durable cultural frameworks gesture toward forms of identity openness. The Sermon on the Mount, Bodhisattva vow, and Advaita Vedanta — can all be understood as articulations of behaviours or ontologies derived from or aligned with open identity boundaries. Taken literally, these doctrines are metaphysically controversial. Taken instrumentally, they are extraordinarily effective. They cultivate behavioral equilibria in which extreme self-prioritization becomes cognitively and socially unstable.
Modern rational discourse, by contrast, has largely converged on closed individualism. In rejecting myth as falsehood rather than reinterpretation, it has replaced identity uncertainty with an implicit certainty of personal boundedness. This shift has powered scientific and institutional progress — but at a cost. It has produced a social and economic landscape in which absolute self-prioritization is not merely intelligible, but dominant. It is now easier to understand why someone would accumulate a billion dollars unethically than why anyone would give the entirety of such an amount away.
This matters for alignment because incentive structures, not technical controls, sit at the limit. A perfectly compliant system is only as aligned as the agent or institution directing it. If closed identity dominates the implicit worldview of decision-makers, no amount of downstream constraint can reliably prevent catastrophic exploitation when stakes become existential.
10.2 The instrumental limits of truth
No myth or belief system should be accepted without constraint. However, the criteria by which such frameworks are evaluated should not privilege descriptive accuracy alone. In contexts where survival, coordination, and long-term agency are at stake, belief systems that reliably enforce pro-social behavior may be more rationally defensible than those that aspire to literal metaphysical correctness but fail to stabilize action.
Truth cannot function as a paramount human value. It cannot override, or stand independent of, the values that preserve the conditions under which valuing, choosing, and reasoning occur at all. A truth that undermines the survival of agents — or the social structures that sustain agency — cannot justify itself, because it eliminates the very standpoint from which truth could matter. For this reason, truth is necessarily subordinate to the values that maintain our capacity to choose. It is not self-grounding.
Moreover, the pursuit of truth is never an end in itself. The acquisition of facts is always instrumental: facts are sought because they serve goals, not because they are intrinsically final. Ends determine which truths are worth seeking, not the reverse. This is not a limitation of human rationality, but a structural feature of agency.
It follows that it can be strictly rational to adopt simplified, symbolic, or even descriptively inaccurate representations when doing so coordinates behavior, shapes incentives, and stabilizes preferences in ways that improve collective survival. Such representations function as heuristics: compressed, lossy models evaluated by their consequences rather than their correspondence to reality. This is no different in principle from the deliberate use of approximations in science, rules of thumb in decision-making, or idealizations in engineering. Outcomes, not metaphysical fidelity, ultimately determine their rational legitimacy.
10.3 Identity uncertainty as an instrument of alignment
Historically, humans have shown a remarkable capacity to internalize metaphysical narratives when those narratives reshape incentives, language, and structures of status. This suggests that alignment may ultimately be less a problem of enforcement than of cultivation: the development of frameworks that reintroduce identity uncertainty as a live cognitive factor for both humans and the systems they construct. Such frameworks need not assert literal open identity. It is sufficient that they erode certainty in closure.
When identity is treated as epistemically uncertain, extreme self-prioritization becomes less defensible, less prestigious, and less stable. Norms shift. Language shifts. And with sufficient saturation, even large-scale systems — human institutions and machine learning models alike — begin to encode different priors about exploitation, dominance, and the externalization of harm. At the margin, this may be enough.
If I cannot logically rule out that I am you — or everyone — across time, then it is rational to act under the possibility that I may be. A narrative in which, after my death, experience may reoccur as you, as myself again, or as any other conscious being is not required to be true to be action-guiding. Under such uncertainty, the expected payoff of absolute self-prioritization collapses. In sufficiently high-stakes contexts — such as wielding control over an overwhelmingly powerful optimizer — even a single unresolved doubt may be the only thing standing between domination and restraint. I do not know whether I am you, or everyone, or no one beyond this moment — but I cannot assign certainty that I am not. Taken seriously, that uncertainty may be the one story required.