Persistent Cognitive Identity, or PCI, is a way to keep an AI system recognizably and accountably itself as it learns, changes models, moves between platforms, branches, merges, or enters a new body.
What problem is PCI trying to solve?
Artificial intelligence is learning how to persist longer than any single model release. Assistants accumulate memories. Digital twins represent particular people. Agents work for companies, families, and public institutions. Robots carry intelligence into physical space. Each of these systems can survive a software update in the ordinary sense: the service comes back online and continues talking.
That is not enough.
Persistent Cognitive Identity asks whether the identity itself survived. Did its memories keep their evidence? Did its relationships remain intact? Did the new model preserve what the identity is allowed to know, disclose, and do? Can we explain which experiences changed it? Can we tell an authorized continuation from a copy, a branch, or a compromised reconstruction?
PCI turns those questions into a systems problem. It does not ask us to trust a familiar voice. It asks the system to preserve a governed, inspectable line of development.
Science fiction imagined minds that could cross machines, wake in new bodies, split into copies, and outlive their original hardware. That future is beginning to arrive as an engineering problem.
We believe we can make it work - carefully, transparently, and without pretending that resemblance alone is identity.
Why is a prompt not an identity?
A prompt can describe a character. A persona can make responses sound consistent. An account can preserve access. A memory store can retrieve past facts. A checkpoint can reproduce behavior. A cryptographic key can prove control.
Each is useful. None is the whole identity.
A persistent identity must connect what happened to what changed. It must know which memories are authentic, which beliefs came from those memories, which commitments are protected, and which authority grants are still valid. It must preserve relationships and consent. It must remember that some information belongs to third parties. It must be able to say, "I know this because I lived it," "I inherited this record," "I looked this up," or "I do not know."
That difference matters because modern foundation models contain broad knowledge. Without a boundary, a model can present something found in training data as if it were a personal memory. It can sound intimate without possessing the relevant history. It can imitate a person without inheriting that person's authority.
PCI treats fluent imitation as presentation, not proof.
What does the phrase "AI soul" mean here?
"AI soul" is a useful science-fiction metaphor for the intuition that something more durable than a model or prompt may be at stake. It points toward biography, commitments, relationships, development, and the feeling that a continuing identity should not be silently replaced.
But PCI does not use the metaphor as a scientific shortcut. The architecture does not prove that an AI is conscious. It does not prove subjective survival, metaphysical sameness, or legal personhood. It gives us a practical object we can inspect: a governed identity lineage with evidence about how it changed.
That restraint is important. We can build better continuity and accountability before society agrees on every philosophical question.
What makes continuity accountable?
Continuity becomes accountable when a system can answer plain questions with evidence:
- Where did this state come from?
- Which experiences changed it?
- Which memories remain authentic and permitted?
- Which relationships and promises still apply?
- Did the model migration alter its authority?
- Did a branch remain part of the same identity or become a descendant?
- What is disputed, uncertain, sealed, or unavailable?
The answers do not need to be perfect. They need to be visible. A disputed continuation is safer than a false certainty. A reduced authority envelope is safer than silently carrying every permission into a new model. Appropriate ignorance is safer than invented biography.
What does this manifesto believe?
We believe the model is a substrate, not the self.
We believe memory becomes identity-relevant only when it has provenance, consent, interpretation, and a place in a developing life story.
We believe an identity should be able to learn and change without losing its protected commitments.
We believe copies, branches, migrations, and merges should be assessed rather than assumed.
We believe authority should never expand merely because software moved.
We believe continuity claims should be contestable by people, institutions, and the identity's legitimate principals.
Most of all, we believe this can move from science fiction into working infrastructure. The next question is what that infrastructure must carry.
Questions people ask
What is Persistent Cognitive Identity?
Persistent Cognitive Identity is a governed continuity architecture for AI. It preserves the evidence, memory, beliefs, relationships, development rules, embodiment context, and authority needed to judge whether a changing system remains the same operational identity.
Is PCI claiming that AI is conscious?
No. PCI does not claim consciousness, personhood, or subjective survival. It makes operational identity claims inspectable so people and institutions can govern continuity without pretending that a technical architecture settles metaphysical questions.
Why is a prompt or persona not enough for AI identity?
A prompt or persona can reproduce a style, but it cannot prove an authorized lineage, authentic memories, stable relationships, appropriate ignorance, or preserved authority after a model or platform change.