The hard ceiling
Gather the threads and the ceiling has a single shape. Few effective channels (Guide 2), discarded fine structure (Guide 2), and current spread (everywhere) together cap how much acoustic information a written percept can carry. That is why music and speech-in-noise have resisted improvement for so long: they are precisely the tasks that need spectral and temporal resolution the electric interface does not deliver.
You can state the ceiling in information terms. If each effective channel is an independent Gaussian channel of bandwidth B and signal-to-noise ratio SNR, the deliverable rate scales with the effective channel count — which current spread pins near a handful. Doubling physical electrodes barely moves N_{\text{eff}}, so it barely moves capacity. This is the quantitative face of 'more electrodes are not more channels', and it borders the broader information-theoretic limits of BCI.
An illustrative capacity bound: total information rate is capped by the small number of independent channels, not by the physical electrode count. Current spread, not wiring, is the binding constraint.
Two ears are still not stereo
Giving a user two implants helps — better sound localization and some noise benefit — but it is not true binaural hearing. Normal listeners exploit microsecond-scale interaural time differences carried by fine structure; independent, unsynchronized bilateral processors transmit those cues poorly, so binaural fine-timing remains largely inaccessible. Where residual low-frequency acoustic hearing survives, electric-acoustic (hybrid) stimulation combines a hearing aid for the lows with electric stimulation for the highs, and often outperforms either alone.
Optics and biology: the credible escapes
The most promising escape from current spread is to stop using current. Light can be confined far more tightly than an electric field, so an optical (optogenetic) cochlear implant could in principle offer many more independent channels — a genuinely higher spectral resolution. The catch is real: it requires delivering a light-sensitive opsin gene to spiral-ganglion neurons, sufficient and safe light power, and chronic stability. Red-shifted opsins and near-infrared/upconversion approaches address power and penetration.
The other credible escape is biology. Preserving and even regenerating spiral-ganglion neurons, and hair-cell regeneration, would improve the substrate every electric or optical device depends on — a device is only as good as the neurons it can recruit. And better central processing, including the attention-decoding front-ends of Guide 4, can extract more from the same impoverished channels.
An honest reading
The auditory field is the one place where a neuroprosthesis is unambiguously, routinely restorative for hundreds of thousands of people — a fact worth stating plainly against the field's frequent overclaiming. But it has plateaued on its hard tasks for a long time, and honest forecasting says the next real gains will come from optics, biology, and smarter processing, not from simply adding electrode contacts against the wall of $1/r$.
One theme unifies this whole track and the volume around it: the read/write asymmetry. Reading the auditory brain (attention decoding, ECAP telemetry to check what was recruited) has advanced faster than writing rich percepts into it. Until we can write with the resolution the cochlea evolved to provide, the deepest limits of the most successful neuroprosthesis will remain limits of writing, not of engineering effort.