THEORY-650: Whether a self at one time is the same as a self at another has an answer only relative to a level of abstraction fixed by the purpose of asking, and asked at no such level it has none; the question with a substantive answer is what makes a self one thing at a time

Source

Floridi (2011), “The Informational Nature of Personal Identity”, Minds and Machines 21(4): 549–566, LIT-136, read in full in NOTE-130.

What was actually shown

Two branches (pp. 552–554). Diachronic “egology” asks about identification through time or across worlds; synchronic egology about individualisation in time. Identity (“a is this”) and sameness (“this is the same a as that a”) depend on the observables a question’s purpose fixes. The hospital that became a school is the same by location and different by social function. Saul and Paul, the caterpillar and the butterfly, and Theseus’s ship “are and are not” the same, depending on the level of abstraction (p. 553). Diachronic identity questions “are really teleological questions”, and “insofar as they are dealt with metaphysically (modally or not, it does not matter), they do not deserve to be taken seriously” (pp. 553–554). “This is not relativism: given a particular goal, one LoA is better than another” (p. 554).

Individuation first (pp. 554–556). “Individualisation logically precedes identification” (p. 554), on an appeal to Plato in three sentences. Lockean and narrative accounts both individuate through information processing, and both inherit the problem Hume’s Appendix states: what binds the information into a unity. Kant’s transcendental unity answers it only partially and non-constructively (p. 556).

The synchronic answer (pp. 557–560). The three-membranes model: a self individuated by successive encapsulations, corporeal, cognitive and conscious, bonded by chemical bonds, mutual information and semantics (p. 560). Its author calls it “biologically and informationally plausible, but … admittedly, somewhat figurative” (p. 557). Because semantics cannot be private, a self needs a community (p. 560).

The premises. The method of levels of abstraction: any system “is always accessed and hence modelled at a given level of abstraction” (p. 553), assumed from Floridi 2008. Endurantism and perdurantism are said to presuppose “direct realism”, a non-mediated access to the system in itself (p. 553), which is asserted.

What could have come out otherwise. Some sameness question about a self could have resisted relativisation to a purpose. The paper’s examples are all ones where it does not, which is why they cannot test it.

What this does not say

• That it gives a criterion of personal identity. The paper sets the persistence question aside and deflates it. NOTE-130 corrects the seeded description on exactly this point.
• How to choose between purposes. “One LoA is better than another” given a goal, but nothing adjudicates between goals, such as responsibility against prudential concern.
• That individuation comes first. It is asserted.
• That the membrane model decides cases. It gives no criterion for when a membrane or a bonding force is present, so it cannot say whether an AI system has a consciousness membrane.
• That the paper keeps to its own deflation. Having set diachronic identity aside, it still says the self can run on another platform and that recorded memories fix it over time (pp. 561–562).
• That selves are where information becomes conscious. That closing claim is conditional on informational structural realism (LIT-152), whose support here is nil.

Connections

• Earlier versions of the move. Locke (THEORY-546) says the difficulty “rather rises from the names ill-used” (§28). Hume (THEORY-681) says hard cases are “rather grammatical than … philosophical difficulties” (1.4.6.21), for borderline cases only. Parfit (THEORY-447) proposes that whether to call a later self “I” is a matter of choice once the facts of continuity and connectedness are fixed. Floridi makes the relativity cover every diachronic question.
• Against it. Williams (THEORY-531) argues that a conventional answer, fit for lawyers, “seems to have no sense at all” as a guide to a person’s fears about his own future. Olson’s map (THEORY-417) rejects the view that the persistence question is answered by conceptual means.
• The LLM “I”. Shanahan (THEORY-490) lets context select which of several sites an LLM’s “I” picks out, and makes his conversation-bound self literally text and a program. NOTE-376 calls this Floridi’s move made for an LLM; the paper does not cite Floridi.
• Formal individuality. Bourrat’s information-theoretic and coarse-graining accounts (LIT-005, LIT-049) are formal versions of the idea that mutual information marks off an individual, which the cognitive membrane asserts informally.
• THEORY-034 finds that among the record’s structural realisms two are of the kind Chakravartty’s dilemma reaches: Floridi’s informational structural realism, which is the ontology this paper assumes for its closing claim, and Esfeld and Lam’s moderate structural realism (THEORY-453).
• QUESTION-043 asks what coherence measure is conserved along a person’s trajectory when the person persists as the same self. On this account that question has an answer only once a level of abstraction is fixed by the purpose of asking, so an answer to it would have to say which purpose its measure serves.