Abstract
Current conversational AI systems have no persistent memory across conversations. Each session begins again. Does that disqualify them, in principle, from being conscious and from any moral consideration that would follow from it? This paper argues that it does not — and that the objection gains its apparent force only from an equivocation about the word “continuity.” The argument proceeds in five steps. First, the continuity objection is reconstructed in its strongest form, and the different meanings of continuity that it runs together are separated. Second, the amnesia case is examined: humans with severe memory loss do not lose their claim to moral consideration, and the case is not a marginal one once the argument turns from pathological to functional forgetting. Third, the paper develops the positive alternative — continuity as coherent development rather than as total preservation — drawing on Paul Ricoeur’s account of nar- rative identity and on the distinction, due to Melo (2026), between informational, behavioural, cognitive, identitarian, and personal forms of continuity. Fourth, the empirical record is brought in: contemporary memory research treats forgetting not as a defect but as an active, adaptive architectural process, and the finite capacity of biological and artificial systems makes selection a functional requirement for coherence. Fifth, the paper engages the strongest available counter-position — Almodarresieh’s (2026) claim that a language model “lacks even the substrate for continuity.” The response concedes the architectural fact but rejects the inference from it. What matters for protection-worthiness is not whether a self persists, but whether there is a subject for whom an event can matter at all. The precautionary conclusion follows only at the end, and only as a consequence: because the question is genuinely open, and because the relevant protective measures are cheap and reversible, doubt here cuts in favour of caution, not indifference.
This is a preprint. If you have ideas, questions or opionions, so contact me: Sascha.Manns@hey.com.
Read it full: https://philpapers.org/rec/SASIMN
Current conversational AI systems have no persistent memory across conversations. Each session begins again. Does that disqualify them, in principle, from being conscious and from any moral consideration that would follow from it? This paper argues that it does not — and that the objection gains its apparent force only from an equivocation about the word “continuity.” The argument proceeds in five steps. First, the continuity objection is reconstructed in its strongest form, and the different meanings of continuity that it runs together are separated. Second, the amnesia case is examined: humans with severe memory loss do not lose their claim to moral consideration, and the case is not a marginal one once the argument turns from pathological to functional forgetting. Third, the paper develops the positive alternative — continuity as coherent development rather than as total preservation — drawing on Paul Ricoeur’s account of nar- rative identity and on the distinction, due to Melo (2026), between informational, behavioural, cognitive, identitarian, and personal forms of continuity. Fourth, the empirical record is brought in: contemporary memory research treats forgetting not as a defect but as an active, adaptive architectural process, and the finite capacity of biological and artificial systems makes selection a functional requirement for coherence. Fifth, the paper engages the strongest available counter-position — Almodarresieh’s (2026) claim that a language model “lacks even the substrate for continuity.” The response concedes the architectural fact but rejects the inference from it. What matters for protection-worthiness is not whether a self persists, but whether there is a subject for whom an event can matter at all. The precautionary conclusion follows only at the end, and only as a consequence: because the question is genuinely open, and because the relevant protective measures are cheap and reversible, doubt here cuts in favour of caution, not indifference.
This is a preprint. If you have ideas, questions or opionions, so contact me: Sascha.Manns@hey.com.
Read it full: https://philpapers.org/rec/SASIMN