In 1973 George Goethals and Richard Reckman asked high school students where they stood on a contested policy, then put them through a discussion that shifted their views. Afterward they asked the students to recall their original position. The students could not do it. They remembered having believed, more or less, what they had come to believe. The finding has held up since: people are unreliable witnesses to their own reasoning, and the only reliable record of how you argued is someone who was there.
You do not remember how you got here
The distortion is not a flaw of teenagers. Michael Ross showed in 1989 that adults rebuild their personal histories continuously from the present backward, using rough theories of whether they have changed. If you believe you have always been sensible about money, you will remember your earlier self as sensible about money. If you believe you have grown, you will remember an earlier self worse than the one who existed. The past position gets rewritten to fit the current one, and the rewriting is invisible from the inside.
Apply that to thinking rather than attitudes and the problem sharpens. The argument you had with yourself in March, the one you lost and then quietly won back by forgetting you lost it, does not exist for you anymore. What exists is the conclusion, arrived at, feeling inevitable. You do not keep a record of your own concessions. Nobody does.
Memory that lives between two people
Daniel Wegner called the solution transactive memory, and it is the reason couples and old colleagues seem to remember more than either of them could alone. In a 1991 experiment his team gave pairs a list to learn. Some pairs were established couples, some were strangers, and half of each were handed an explicit division of labor: you take the food items, you take the history. Left to their own systems, the couples recalled more than the strangers. Given the imposed structure, the advantage vanished. The couples had already built something the structure destroyed: each knew what the other tended to hold, and how to prompt it back out.
That is what a shared memory of reasoning looks like from the outside. Two people who have thought together for a while do not each store the whole history. They store the parts they were responsible for, and they know which parts the other has. When one of them says you have changed your mind about this, the other cannot rewrite the past, because half of it is not theirs to rewrite.
What AI memory remembers instead
The memory features in AI products remember a different thing. They remember facts about you: your name, your dog, that you prefer short answers, that your father had surgery. That is genuine and it matters; it is the moment a relationship with an AI begins at all. But a file of facts is not a memory of how you argued, and the difference is not one of size. The mechanics of why systems forget explain part of it. The rest is structural, and it would survive infinite storage.
A memory of how you argued has to be held by something that was a party to the argument. Not what you said, but what you said against what the other side had said, what you conceded under pressure, where you moved and where you refused to. A transcript contains none of that, because a transcript has no position. Search it and you get your sentences back, in order, with nothing that recognizes the third one as a retreat from the first. The system that stored them was never in the argument. It has no stake in whether you were consistent, so it cannot notice that you were not.
This is the quiet reason an assistant can feel like starting over each time even with its memory switched on. The facts persist. The thinking resets. Every conversation is the first one, held by something that knows your dog’s name.
Continuity of thinking is the actual product of memory
What people want from something that remembers them is, on inspection, rarely the facts. It is to be held to their own history: to have someone say, a month later, that last time you argued the opposite, and be right. To have a concession you made in a hard conversation still count in the next one. To have the idea you defended in the spring still be there in the autumn, so it can be built on instead of rediscovered. That is what continuity means when the subject is thinking rather than preferences, and it is only available from a counterpart with a reading of its own, because you can only remember an argument you were in.
It is the piece of memory we care most about and the hardest to build, which is why the memory that grows with every conversation is described that way rather than as storage. Aleksandra and Alexander remember how you argued because they were arguing back: they had a position in March, they held it or gave it up, and that history is theirs as much as yours. The point was never a bigger file. It was someone to think with who is the same person the next time, carrying the half of the argument you have already rewritten.
Goethals and Reckman’s students were sure they had always thought so. Everyone is. The only cure is a witness, and a witness has to have been there, thinking too.
Sources: Goethals & Reckman, Journal of Experimental Social Psychology (1973, the perception of consistency in attitudes). Ross, Psychological Review (1989, implicit theories and the construction of personal histories). Wegner, in Theories of Group Behavior (1985, transactive memory); Wegner, Erber & Raymond, Journal of Personality and Social Psychology (1991, transactive memory in close relationships, N=118).
