Persistent context
Memory correctability
Memory correctability is the property of an AI system whose retained picture of a person can be inspected and changed by that person. It requires that the record be legible in the first place: a memory nobody can see is a memory nobody can correct.
Also called editable AI memory, memory legibility
Systems that retain personal context get things wrong in a specific way. They do not merely store a false fact; they store an inference, drawn from something a person said once, in a mood, about a situation that has since changed. The inference then quietly conditions everything that follows, and the person experiences it as the system being subtly, persistently off.
Correctability treats that as a first-class failure and gives it a remedy. It means a person can ask what the system thinks it knows, get an answer in terms they can actually evaluate, and change it. That is harder than an edit button, because much of what a memory system holds is not stated in sentences a person would recognize.
There is a design tension here worth naming. A memory that is fully exposed can feel like a surveillance report, and reading it can be unpleasant even when nothing in it is wrong. Legibility has to be built so that inspecting the record is a normal act, not an alarming one.
Why it matters
Without correctability, every mistaken inference a system makes about a person is permanent and invisible. The person is left managing a version of themselves they cannot reach, and their only remedy is to abandon the system entirely.
Where TAICU stands
TAICU treats legibility and correctability as prerequisites for retaining personal context at all, rather than as controls added once the memory system already works.
Common questions
- Why can’t people just delete their AI’s memory?
- Deletion is available in most systems, but it is a blunt remedy: it costs the person all the continuity that made the system worth using. Correctability is the finer-grained ability to fix what is wrong while keeping what is right.
- What makes an AI memory legible?
- That a person can ask what the system believes about them and receive an answer they can evaluate — in their own terms, not as an opaque score or an embedding — and then change it.