Drugs that affect the nervous system get administered systemically. It's easy to imagine that we could do much more if we could stimulate one nerve at a time, and in patterns designed to have particular effects on the body.
"Neural coding" can detect the nerve impulses that indicate that a paralyzed person intends to move a limb, and build prosthetics that respond to the mind the way a real limb would. A company called BrainGate is already making these. You can see a paralyzed person using a robotic arm with her mind here.
A fair number of diseases that don't seem "neurological", like rheumatoid arthritis and ulcerative colitis, can actually be treated by stimulating the vagus nerve. The nervous system is tightly associated with the immune and endocrine systems, which is probably why autoimmune diseases are so associated with psychiatric comorbidities; it also means that the nervous system might be an angle towards treating autoimmune diseases. There is a "cholinergic anti-inflammatory pathway", involving the vagus nerve, which inactivates macrophages when they're exposed to the neurotransmitter acetylcholine, and thus lessens the immune response. Turning this pathway on electronically is thus a prospective treatment for autoimmune or inflammatory diseases. Vagus nerve stimulation has been tested and found successful in rheumatoid arthritis patients, in rat models of inflammatory bowel disease, and in dog experiments on chronic heart failure; vagus nerve activity mediates pancreatitis in mice; and vagus nerve stimulation attenuates the inflammatory response (cytokine release and shock) to the bacterial poison endotoxin.
We'd need much more detailed maps of where exactly nerves innervate various organs and which neurotransmitters they use; we'd need to record patterns of neural activity to detect which nerve signals modulate which diseases and experimentally determine causal relationships between neural signals and organ functions; we'd need to build small electronic interfaces (cuffs and chips) for use on peripheral nerves; we'd need lots of improvements in small-scale and non-invasive sensor technology (optogenetics, neural dust, ultrasound and electromagnetic imaging); and we'd need better tools for real-time, quantitative measurements of hormone and neurotransmitter release from nerves and organs.
A lot of this seems to clearly need hardware and software engineers, and signal-processing/image-processing/machine-learning people, in addition to traditional biologists and doctors. In the general case, neural modulation of organ function is Big Science in the way brain mapping or genomics is. You need to know where the nerves are, and what they're doing, in real time. This is likely going to need specialized software which outpaces what labs are currently capable of.
Bioelectronics seems potentially important not just for disease treatment today, but for more speculative goals like brain uploads or intelligence enhancement. It's a locally useful step along the path of understanding what the brain is actually doing, at a finer-grained level than the connectome alone can indicate, which may very well be relevant to AI.
It's tricky for non-academic software people (like myself and many LessWrong readers) to get involved in biomedical technology, but I predict that this is going to be one of the opportunities that needs us most, and if you're interested, it's worth watching this space to see when it gets out of the stage of university labs and DARPA projects and into commercialization.