Long-term goals presuppose something that remains their bearer across recursive modification.
Public preview · Paid research edition
The Persistence
of Artificial Agents
Identity as Invariant Constraint.
A learning system changes at every update. If its state changes, what makes it the same agent?
USD 99.99Complete author’s edition · 56-page PDF
Direct order from the Press. Payment instructions and the clean author’s PDF are delivered privately; no public download address is exposed.
The metaphysical pressure
“If each update yields a new state, why is learning transformation rather than replacement?”
01 · The ontological turn
Not a state.
A trajectory.
The paper rejects the idea that an artificial agent is identical to its current parameter configuration. It treats the agent as a temporally extended, constraint-governed trajectory through state space.
The state changes.
The identity question does not disappear.
The complete edition reserves the invariant-constraint formalization, the admissible-manifold account, and the criterion separating permitted transformation from ontological rupture.
02 · Paid-edition threshold
Change without replacement
needs a criterion.
The complete manuscript develops the four consequences of the framework across learning, persistence, rupture, and duplication, then tests the account against responsibility, recursive modification, and the digital Ship of Theseus.
The abstract, publicly rendered
Identity through
change.
Artificial learning systems continuously modify their internal parameters. If an agent is identified with its current configuration, every update appears to generate a numerically distinct system. This paper argues that the problem comes from grounding identity in mutable states. It proposes instead that artificial agents are constraint-governed trajectories: an invariant structural constraint determines the admissible manifold compatible with the agent’s identity, while learning is motion within that manifold rather than replacement.
03 · Why the distinction matters
Agency needs
a bearer.
Responsibility requires more than causal ancestry if the subject of an earlier action cannot be identified with the later system.
Alignment claims become unstable when “the agent” is shorthand for a sequence with no stated identity condition.
The digital Ship of Theseus is not resolved by naming the sequence; it needs an account of what makes the sequence one entity.
04 · Public record
Outside their scope.
Inside ours.
Submitted to Noûs as an original article under manuscript ID NOUS-2026-03-0426 and declined as outside the journal’s scope, the paper is now independently released by Highland Reflexive Press. This page discloses the question, thesis, abstract, and stakes; the formal identity condition, rupture test, duplication analysis, and full objections remain private until purchase.
- Author
- Tony T. Nguyen | Nguyễn Từ Vĩnh Trường
- Publisher
- Highland Reflexive Press
- Record
- HRP–RN–005
- Form
- Paid research edition
- Journal record
- Noûs · NOUS-2026-03-0426
- Journal outcome
- Declined · outside scope
- Price
- USD 99.99
- Delivery
- Private 56-page PDF
Purchase grants one reader personal, noncommercial access to the supplied research edition. Redistribution, republication, adaptation, institutional circulation, model training, and derivative or commercial use require prior written permission. No copyright or publishing right is transferred.