What Does It Mean for a System to Still Be Itself?
Across identity systems, artificial intelligence and institutions, we keep encountering versions of the same practical problem.
Something can continue producing the correct signal after the conditions that originally made the signal meaningful have changed.
A private key still signs. An account still authenticates. A procedure is followed correctly. An AI produces an answer that sounds appropriate. None of those observations necessarily tells us whether the larger situation remains what we think it is.
This distinction becomes especially visible under stress.
The vault is empty, but there's an insurance claim. The data is inaccessible, but there's a ticket number. The account is frozen, but an appeal is pending. These aren't necessarily examples of malfunction. The systems may be behaving exactly as designed. What has changed is the question we're asking.
We no longer want to know whether a valid claim exists.
We want to know what exists now.
That difference between the signal and whatever the signal refers to has occupied a great deal of my thinking about infrastructure, identity and long-running interaction. I previously tried to gather those observations under a general theory of coherence.
I think I was moving too quickly.
There is still something here worth measuring. I'm less certain that all of these domains contain the same thing.
Control is not continuity
Cryptography gives us unusually precise tools for establishing control. If a valid signature can only be produced with access to a particular private key, possession of the required key tells us something important.
It doesn't tell us everything.
Suppose an organization has used the same multi-signature arrangement for ten years. Every signature remains cryptographically valid, but the people have changed, the institution has changed, its purposes have changed and perhaps the relationships that originally justified its governance structure have changed too.
The cryptography hasn't failed.
We've changed levels of description.
The same problem appears in less technical forms. A newspaper can retain its name while changing ownership, staff, audience and editorial practices. A town can retain its legal boundaries while becoming a substantially different community. A person can remain legally and biologically continuous through enormous changes in belief, ability, relationships and behavior.
We still use the word same in all three cases.
That doesn't mean sameness is the same property in all three.
The useful question is more specific: what are we trying to preserve, and what evidence would tell us whether it survived?
Persistence isn't identity either
My earlier answer was coherence.
I defined coherence through continuity, reversibility, non-contradiction, persistence across interruption and non-degenerate interaction, then treated those conditions as an invariant capable of explaining identity across different kinds of systems.
That was elegant.
Probably too elegant.
Each proposed property may still be useful, but none deserves universal status simply because it travels well between examples.
Continuity matters when history matters. Reversibility matters when repair matters. Contradiction matters when incompatible commitments need to be recognized. Persistence matters when state has to survive interruption. Variation matters when adaptation or genuine exchange requires alternatives.
Different systems need different combinations.
A database can be extremely useful while being terrible at revisiting its own assumptions. A legal institution may deliberately make some decisions difficult to reverse because other people need to rely on them. A conversation should tolerate correction in ways we wouldn't want from a financial ledger.
Even contradiction requires care. New evidence can legitimately overturn an earlier conclusion. Two statements that appear inconsistent may apply under different cond