Source
Lewis (1976), “Survival and Identity”, in Rorty (ed.), The Identities of Persons, pp. 17–40, LIT-446, read in full in NOTE-373. Postscript A of the 1983 reprint was read beside it and is not the paper’s text.
What was actually shown
Two answers, both kept. “What matters in survival is mental continuity and connectedness” (p. 17), and “what matters in survival is survival”, identity of the I now and the surviving I (p. 18). If they disagreed, “we would have to prefer the platitude of common sense” (p. 18).
Parfit’s argument, with its premises granted. (a) Identity is one-one and not a matter of degree; (b) mental continuity and connectedness need not be; so (c) what matters is not identity (p. 19). Lewis grants (a) and (b) and refuses to redefine “identity” (p. 20).
Different relata (pp. 20–23). Continuity and connectedness hold among “more or less momentary person-stages”; identity among “temporally extended continuant persons” (p. 20). “Even if you survive, your present stage is not identical to any future stage.” The fair comparison is between R and the I-relation, which holds between two stages when some one continuant person has both (pp. 21, 23). If I is R, “something is a continuant person if and only if it is a maximal R-interrelated aggregate of person-stages” (p. 22). The argument that I must share identity’s form “wrongly takes it for granted that every person-stage is a stage of one and only one continuant person” (p. 23).
Fission and fusion as overlap (pp. 24–25). The I-relation fails to be transitive “if and only if there is partial overlap among continuant persons”. In fission two continuants share their pre-fission stages.
Counting (pp. 26–29). With overlap there are more continuants than stages at a time, so we count stages: continuants are identical-at-t when their stages at t are identical (p. 26). “Tensed identity is not a kind of identity” (p. 27). Two roads that merge for a stretch are the model, and a name for someone who will divide is harmlessly ambiguous until the division (pp. 27–29).
Degrees (pp. 32–36). Naive degrees fail: an aggregate can pass the R-test to degree 0.9 and fail maximality to degree 0.88, and so be “a person only to degree 0.02” (p. 34). Treating each cut-off on R as a delineation, the measure of delineations on which two stages are I-related equals the measure on which they are R-related (pp. 35–36).
What could have come out otherwise. The I-relation could have been shown to inherit identity’s form whatever persons are. It does not, because nothing forces each stage to belong to one person only.
What this does not say
- That persons are stages and their sums. That is assumed (n. 4), and defended only in Postscript B (1983).
- That I equals R in every case. It holds by definition once persons are the maximal R-interrelated aggregates. That definition is the stipulation, and “at least, I cannot think of any that clearly is not” (p. 22) is its only support.
- That the cost is small. Lewis calls counting by tensed identity “somewhat counterintuitive” (p. 38), and on the Methuselah case continuum many persons share a birthday and are counted as one (pp. 29–31).
- What a stage that will divide should fear or hope. The paper does not ask. Postscript A’s plural desire says what it can want, not how it should feel about the branch that will soon die.
- That the answer is stable under a choice of R. The argument works for any version of R (p. 20), but which aggregates are persons depends on the version through maximality.
- Anything about the first-person argument from expectation. Williams’s case (THEORY-531) has no branching, so tensed identity gives it nothing to do, and the paper does not discuss it.
- That a recording played back into a new body is survival. That is Postscript A (1983), from Leiber’s novel, not the paper.
Connections
- Rival. Parfit (THEORY-447). Lewis accepts that paper’s account of what matters and its premises about identity, and denies only that what matters is not identity. Olson’s survey (LIT-454) files this as the “multiple occupancy” horn of the fission dilemma, with Perry; Lewis’s persons are “branches” where Perry’s are “lifetimes” (pp. 37–39).
- Counting overlapping individuals. Meincke’s pregnancy (THEORY-448) is the same problem in biology. Lewis keeps identity sharp and lets persons share parts; Meincke gives up substance counting for a count between one and two. Neither engages the other.
- Machine instances. The LIT entry observes that stage-sharing lets instances of one model that share their earlier stages be distinct continuants counted as one until they diverge, beside Goldstein and Lederman’s instance agents (LIT-206, NOTE-138). That application is the filing reader’s, not Lewis’s. Shanahan’s context-relative “I” (THEORY-490) treats the same multiplicity by usage, not by counting.
Note of 2026-10-30: what holds it back
Kept Proposed by the twenty-fourth status review. The title is a conditional in the record’s voice, not a report of Lewis. Its middle clause, “so Parfit’s premises do not show that identity is not what matters”, is Lewis’s interpretive conclusion against THEORY-447, a declared rival. Whether it is fair to read “identity matters” as the I-relation among stages is the point in dispute (promotion-audit decision of 2026-10-20). What is missing is the second half of the condition: a showing that what a person wants in wanting to survive is something a stage can want. The condition’s first half asks for an argument for the antecedent. A conditional title does not assert the antecedent, so that half asks for more than the title (ADR-065 §1). It is left as it stands.