By "high level" I mean replacing macro-scale brain structures with hardware
based on a foreign computational architectures.
Actually, the difficulty isn't so much in the hardware as it is in the
interfacing software. On a micro scale this is fairly easy since you are
only looking at discrete signals. On a macro scale, however, you also have
to consider the relationships between signals on the order of millions or
billions of simultaneous signals. This has to be done both going in to the
hardware and going back in to the brain. The computational complexity of
decoding this type of interface would require immense quantities of
computational power.
You could use neural network based enhancement hardware to minimize the
interfacing problem, but neural networks are not efficient computational
architectures for many classes of problems. In the long run, it would break
down to how much you are willing to spend for computational enhancement.
The cost/performance ratio could get prohibitively large very fast, based
almost exclusively on the complexity of installation and integration.
-James Rogers
jamesr@best.com