New paper links neural network gradients to the structure of phenomenal experience

The paper argues that these Jacobian-based measures can account for various qualitative aspects of consciousness. Specifically, the model aims to explain phenomena such as the duration of experience, which can persist over hundreds of milliseconds, and the distinction between vivid and obscure perceptions. It also addresses the experience of texture, the "blooming buzzing confusion" associated with newborns, and the difference between distinct and confused mental states. Furthermore, the work seeks to clarify what learning feels like and the functional role of rich, dense experience.

This research sits at the intersection of neuroscience, philosophy of mind, and machine learning. By formalizing the link between gradient structures and subjective experience, the authors provide a testable hypothesis for how computational systems might generate or mimic conscious states. The approach offers a new perspective on the hard problem of consciousness by grounding it in the mathematical properties of differentiable systems.