[pharke]

I would argue that that number is a bit high for sense input. Figures I've seen are about 20x lower on the high end. The only time I've seen estimates as high as the one you gave are when they are considering the full field of vision for each eye and translating that directly to pixels assuming there is 100% fidelity. Human vision doesn't work that way, we dedicate far fewer neurons to our peripheral vision outside the fovea. You also have to remember that neurons transmit data at a much slower rate than electrical connections (100m/s vs 300,000km/s) so the path to building a better brain machine interfce is to have a large number of electrodes which seems to me a fairly straightforward problem to solve e.g. make them smaller and adding more of them. The other side of the coin is that we aren't aware of all of our sensory input all of the time so you can get away with only transmitting the important stuff which greatly reduces the bandwidth needed. Further out, there's likely some form of abstraction going on inside our brains and if we can communicate data on that level we'll likely need even less input to create a realistic experience.

As an aside, I don't think I'd opt for the type of uploading outlined in that novel considering it involves a posthumous destructive scan of the connectome. There wouldn't be any continuity so you'd probably just end up with a copy. The only type of uploading I could see working would be a Ship of Theseus approach where you gradually replace your biological neurons with artificial neurons and even that causes some hesitation.