A quick look at the motor control papers in the 1981 handbook of physiology
(all on Smith & Smith's bad guy list) makes it seem to me that their authors
are streets ahead of the `compute the inverse dynamics' crowd, going for
complicated and messy mixed models where anticipation & feedback are
combined, etc. So far, I'm quite puzzled as to how Jordan & Rosenbaum's
chapter got into a cognitive science book, since anybody who had skimmed the
handbook of physiology chapters would have at least thought of the possibility
that the spinal reflex loops might soak up some of the computational
burden. Perhaps a failure to distinguish between routine activity
and performance at the limits is striking again.
So far, I'm quite puzzled as to how Jordan & Rosenbaum's
chapter got into a cognitive science book, since anybody who had skimmed the
handbook of physiology chapters would have at least thought of the
possibility that the spinal reflex loops might soak up some of the
computational burden.
Not anybody: odd as it may seem to those of us to whom this kind of
thinking is natural, that "the spinal reflex loops might soak up some of
the computational burden" is actually a pretty sophisticated idea
involving (at least implicitly) layers of virtual machinery and
distributed processing, and takes more knowledge of computing than is
possessed by many CS undergrads. In fact I suspect the modern tendency
for CS undergrads, and computer users in general, to live in the cosy
protected computational worlds simulated by high-level computing
environments makes this kind of idea even harder to grasp.
PCT's difficulty is that understanding it needs a mix of cognitive
skills that almost no formal educational program running today provides.