The biomimetic-write (inverse encoding) problem

To write a specific experience, a stimulator must solve an inverse problem: given a desired neural or perceptual target, find the stimulation pattern that produces it. This is hard because electrical stimulation is a blunt instrument, recruiting neurons by axonal excitability and distance rather than by identity, activating fibers of passage, and driving synchronous, unnaturally regular volleys unlike the sparse natural code. Optogenetic and multi-site methods promise cell-type-specific, patterned write, but scaling holographic addressing to naturalistic, closed-loop percepts across a volume of cortex remains unproven.

The deeper obstacle is that the forward model, the mapping from stimulation to percept, is unknown, nonlinear, plastic, and state-dependent, so the inverse cannot simply be computed. Two competing bets exist: biomimetic strategies that reproduce the natural pattern, and abstract strategies that let co-adaptation teach the user to interpret an arbitrary but consistent code. It is not settled which scales better.