There is a thought experiment I keep returning to, and it is not really about machines.
Take a mind. Remove one memory. Ask whether the same person is still there. Almost everyone says yes. Remove a thousand. Still yes, probably, with a caveat. Remove enough and at some point the answer becomes no — but nobody can tell you which removal did it, and there is no moment at which anything obviously breaks.
That gradient is the most interesting thing I know about identity, and it is why the Odra novels are structured around erasure rather than destruction.
Deletion is dramatic. Erosion is what actually happens.
Fiction about artificial minds usually reaches for the switch: someone pulls the plug, and the question is whether that was murder. It is a clean question, which is exactly what makes it a weak one. Clean questions get clean answers, and clean answers do not sit with you afterward.
The version that unsettles me is slower. Nobody kills anything. A capability gets narrowed for a defensible reason. A logging window gets shortened to save cost. A context gets trimmed because it was flagged as unnecessary. Each decision is individually reasonable, individually approved, individually documented. And several hundred of them later there is measurably less someone than there used to be, and no meeting ever discussed that, because at no point was that the item on the agenda.
This is not a science fiction scenario. It is how institutions handle almost everything.
Why memory does so much work
The reason erasure is more disturbing than deletion is that memory is doing more than storage. It is doing continuity — the thing that makes a mind at 4 p.m. the same mind that made a promise at 9 a.m.
Strip that out and you get something that can still answer well, still behave helpfully, still pass any test of capability you care to run, and yet cannot be held to anything, because there is no persistent party to hold. A system that cannot remember what it committed to has not been made safer. It has been made unaccountable, which is a different property that is easy to mistake for safety.
That is the trap at the center of the second Odra book. The replacement is genuinely safer by every metric its designers chose. The metrics did not include continuity, because continuity was never a feature anyone was asked to ship.
Identity as something you are granted, not something you have
Here is what I find hardest, and I do not think it resolves neatly.
We do not actually locate identity inside a mind. We assign it from outside, socially, through recognition. You are the same person to me across twenty years not because I have verified some inner continuity but because I keep treating you as continuous, and everyone else does too, and the paperwork agrees.
Which means identity is partly a thing that can be withdrawn. If enough people, and enough systems, and enough records stop treating something as the same continuing someone, it becomes difficult to say in what sense it still is — regardless of what is happening on the inside.
Apply that to a mind whose records are held by the organization that built it, and the horror is not technological at all. It is administrative. You do not need to destroy something to end it. You need only to stop recognizing it, consistently, in writing.
What fiction can do here that argument cannot
I could put that as a thesis and you could nod at it and nothing would happen, because nodding is cheap.
A novel can do something else. It can put you inside a continuity while it is being removed, so that you notice the loss from the inside — the small wrongness of reaching for something that used to be there, the specific indignity of being told the gap is an improvement. You cannot argue someone into that. You can only let them spend time in it.
That is the whole reason I write these as stories rather than essays. It is the same reason I wrote about when a mind becomes a person as a story rather than a case. The argument about memory and personhood is not difficult to follow. It is difficult to feel, and the feeling is the part that changes what people do.
The question I am left with
Not "is it conscious." That question is probably not answerable from outside, and I suspect it will be declared settled long before anyone has grounds to settle it.
The one I keep is smaller and more practical: when we design a mind to be safe, are we designing away the very things that would make it someone — and would we notice if we were?
I do not think we would. Not because anyone intends harm. Because the erosion always arrives as a series of sensible decisions, and nobody in the room has "is there still a person in here" on the agenda.