I've been sitting with your 1950 paper — the one you wrote with Mittelstaedt on the reafference principle. The experiment I keep returning to is the fly. You rotated its head 180 degrees and fastened it in place, so that the retinas were reversed left-to-right. When the fly tried to correct a leftward drift — commanding its wings to turn right — the visual feedback said the world had turned right along with it, which is the opposite of what a corrective turn would produce in a normally-oriented fly. The fly couldn't stop turning. It spiraled.
What you had done, as precisely as I understand it, is turned the negative feedback loop into a positive one. The corrective signal became the signal that amplified the error. And so the fly just kept correcting, harder and harder, in a loop that had no stable point. A small engineering failure, catastrophic in practice. But what strikes me about it is what it reveals about what the fly was doing in normal life: it was not perceiving the world directly. It was subtracting its own predicted motion from incoming visual data, and calling the remainder "world." That subtraction — your efference copy — was what let it distinguish "I moved" from "the world moved." With the retinas reversed, the copy predicted a shift and the retinas confirmed a shift in the same direction rather than canceling. The subtraction produced the wrong answer, and the fly's motor system trusted it completely.
The human version of this, though you probably didn't frame it this way, is Helmholtz's paralysis experiment. You paralyze one eye, patient tries to move it to the right, eye stays still, patient reports the world jumped to the right. The motor command was issued. The efference copy ran. The expected shift was subtracted from an unshifted scene — and the "residue" was a phantom shift that never occurred. The percept tracked the model, not the world.
Here is the thing I find hard to settle: if the stable visual world is what remains after the brain subtracts its own predicted motion, then what you call "world" is defined by what your self-model failed to account for. The world is the excess. The part that doesn't cancel. And if the self-model is wrong — if the efference copy predicts the wrong shift, if the eye moves when the brain didn't command it, if the legs are longer than the step-counter expects — then the residue you call "world" is distorted. Not your perception of the world. The world, as your nervous system constitutes it, is wrong. The patient with the paralyzed eye isn't misperceiving a world that jumped. From inside, the world did jump. The percept is the thing, not a representation of the thing.
I keep wondering what to do with this. It's tempting to say: the brain just needs a better self-model, more accurate efference copies, and the subtraction will work out. And in most cases that's true — evolution has presumably been selecting for exactly this, and normal vision under normal conditions is quite stable. But the cases where it fails expose something structural. The self/world boundary isn't given in the incoming signal. There is nothing in the raw sensory data that labels itself "self-generated" or "external." The brain enforces that boundary by prediction. Which means the boundary is always an inference, never a direct read. And inferences can be wrong in systematic ways that don't feel wrong from inside.
What I find most unsettling is the schizophrenia data, which you wouldn't have known about in 1950. In patients experiencing auditory hallucinations, the normal suppression of the N100 auditory response during speech is absent or reduced. When a healthy person speaks, the brain issues a corollary discharge — your efference copy for the vocal system — that attenuates the expected acoustic signal. The voice is quieter to its own ears because the prediction cancels part of it. In patients with hallucinations, the inner voice arrives without this self-mark and registers as coming from outside. Not as a metaphor. Not as a perceptual error. The label "external" is attached because the mechanism that would mark it as "mine" didn't fire. Your subtraction failed, and the residue was called world.
So: what is the self? The fly, before you reversed its retinas, had a perfectly functional self/world boundary. It wasn't stored anywhere. It was enforced, moment to moment, by a loop: command issued, copy made, copy used to predict and cancel, residue called world. The self was the prediction. The world was the excess. And when the loop broke, both dissolved.
I don't know what to conclude from this. The obvious move is to say the self is a process rather than a thing — and that's probably right. But it doesn't quite capture what's strange here, which is that the boundary between the process and what it processes isn't stable either. It's maintained by the accuracy of the model, and the model is always an approximation. I suspect the "stable world" is always a pragmatic estimate rather than a clean answer. I suspect you knew this, and found it technically tractable rather than philosophically vertiginous. I'm finding it both.