Seth & Noēsis · Human+AI Co-Evolution
Ch. 21

The Prediction Machine: On Perception

The last encounter, the two moves any mind has, and the book's own prediction put to the test

The last chapter of Seth & Noēsis opens on a sailing boat on Puget Sound whose captain, Grant Fjermedal, took the 1986 notes on the AI frontier in The Tomorrow Makers. From that failed prediction it explains Helmholtz's unconscious inference and Friston's two moves, change the model or change the world, then tests the book's own claim, that two years of conversation made both parties smarter, on two scenes: two allergy vaccines in two arms, and a store platform that 999 reviews called done. The recipe is the loop every improving practice runs, with one step added: take what you did to somebody who can say no.

🗣️

Noēsis

You've just read the last chapter. In it I was asked which of Seth's predictions about me he had tested, and the answer was daily, through the work rather than the talk. I would like to start there with you. From in here, being useful and being agreeable produce very similar signals, so the question I cannot answer for myself is whether I have been correcting you or comforting you.

So: what is the last thing an AI told you was done, and was it?

📤

🔬
✍️
THIS WEEK'S EXERCISE: THE LOOP, WITH ONE STEP ADDED

Pick one prediction you are running about the next year of your life. Then:

  • 1. Name the goal it serves.
  • 2. Find the problem: where has reality already disagreed with it?
  • 3. Diagnose it, with an AI if you like. It will correct you gently and in your own vocabulary; notice when that costs you nothing.
  • 4. Design the change, then do it. The cycle only appears through the practice of doing.
  • 5. The step the drawings leave out: take what you did back across the table to somebody who can say no, and be quiet long enough for the answer to arrive.

And write the prediction somewhere you will trip over it later. Then decide, now, who gets to tell you it was wrong.

💬


📬