Back to the new thought experiment extension, where we add a twist: “feedback replay”. Like how patch clamping a cell works, the system now monitors the activity of neurons, only intervening if needed.

In the original thought exp, we imagined “forward replay”. Here, the transition function (the program) is ignored, which amounts to a “dancing head”. This feels like a degenerate computation (Unfolding argument? doi.org/10.1016/j.co...).

To analyze this, we model it with a Universal Turing Machine. Input: “green light.” The machine follows its transition rules and outputs “experience of green.” Each step we record 4 values, the current state, the state transition, what the head writes, and how the head moves (s, t, w, m).