The Neuralink presentation a few days ago made the important point that the physics of neurons makes it impossible to read and/or write their state with any kin

The Neuralink presentation a few days ago made the important point that the physics of neurons makes it impossible to read and/or write their state with any kind of accuracy without getting very, very close to them. As such, these kinds of completely non-invasive methods of interfacing with the brain are a dead end. Barring breakthroughs in scanning technologies that completely upend basic laws of physics, you will always be limited to a low-bandwidth channel that only works by reading the crude, aggregate state of the whole brain or a large area of it, and requires extensive training for participants to learn how to send basic signals. Since the limits here are hard physical ones, not ones that can be engineered around, there's no way for this to be gradually refined into a more useful system. It will ways be a hack. If we're going to produce high-bandwidth brain-machine / brain-machine-brain interfaces that allow useful collaboration, it's going to require getting inside the skull and getting up close and personal with brain matter, whether we like it or not.