[Avery.Andrews 930911.1519]
Thinking about Fowler & Turvey (1978) is causing me to propose a new
blunder to add to Bill's list, which I'll call the `KR (knowledge of results)
Blunder' (though it's really a muddle). The KR Blunder (highly
evident in Jack Adam's `A Closed Loop Theory of Motor Learning') is the
failure to distingish clearly between the problems of explaining skill
acquisition on the one hand, and skill implementation, on the other.
Consider hitting a baseball. There are two possible conceptions of what
this skill involves:
a) the learner develops a Central Pattern Generator (CPG) that
takes as input info about the trajectory of the approach ball,
and produces as output a contour of reference levels for joint angles,
such that ....
b) the learner acquires a control system that controls for a complex
relationship between the location & movement of the ball and various
joint angles, such that ...
... at the appropriate the time, the bat connects with the ball in the
appropriate manner.
PCT'ers presumably would favor (b) over (a), but I'd be happy to let
the choice rest on the outcome of empirical work (I'm actually not
entirely convinced that the two are empirically distinct).
What is clear, however, is that the skill, once acquired, consists of
setting the right reference levels for lower level control systems
(Gary's experiments on throwing a tennis ball with an elastic band
attached to your arm are an excellent demonstration of this).
What `closed loop' theories of motor learning say is that acquisition
of a skill involves Knowledge of the Results of attempted actions in order
to tune the CPG or whatever that's producing the action. The KR Blunder
is then to fail to distinguish the role of KR in skill acquisition, from what
one might call `Knowledge of Status' (KS) in skill implementation
(the relevant status being r-p, the difference between how things are
and how they are wanted to be). At a deeper level of analysis, KR is
an instance of KS (what your current organization produces as opposed
to what you want it to produce), but they are KS for different control
systems, which are not distinguished by Adams or F&T (I'd like to check
out N. Bernstein on this, but he's pretty heavy going, and there are
time constraints).
F&T is ostensibly about skill acquisition, but what they criticise is
an early PCT theory of skill implementation (and their presentation indicates
that they didn't really understand how it is supposed to work, for
example when they say that you consult memory to figure out what
correction to produce in response to what error signal). Their own task,
guiding a cursor to a point using a complicated function of handle-position
to cursor location, is actually not overtly an instance of skill acquisition
at all--the subjects simply perform a task using a trial-and-error method
easily simulated by PCT models with reorganizational capacity. It is
however a reasonable preliminary sketch of how a kind of reorganization
system might work, similar in nature to the E. Coli demo (so I was wrong
yesterday when I said that the paper had no merits at all).
Their confusion between these two levels is particularly
unfortunate, because it seems to me that they could have made some
apt criticisms of the BCP 2nd order systems as a theory of skill
implementation. For example, BCP assumes that 2nd order perceptions
are linear combinations of first order ones: if this is true, and all
of the first order perceptions have first order control systems to
control them, then you actually don't need second order *control* systems
at all-you can do calculations to find a combination of first order
perceptual signals that will yield the desired combination of second
order ones (if there is one), and let the first order control systems
do the actual controlling.
On the other hand, if the second order perceptual functions are
nonlinear, there won't in general be any fixed way to map the second
order error signals onto first order reference signals that will
result in control (I think F&T might have been trying to say this, if
they actually had done so, it would have been a much better paper).
To see this, consider a nonlinear 2nd order perceptual function z
of two variables x and y. Z is presumably continuous, so it is basically
a landscape with no caves or overhangs, with the xy plane for z=0 as
horizontal (`sea-level', let's say). The current values for x_p and y_p
(perceptions), determine a point on the landscape with a particular
elevation z_p, but then z_r shifts to a new elevation. The problem
is to find a followable attainable course values for x_r and y_r that
will lead x_p and y_p to a point where z_p equals z_r. This can be
done by `hill-climbing' techniques, plus random jumping when you get
stuck on a hilltop or a basin), but this isn't skilled performance.
(So F&T don't actually model skill acquisition, but rather a kind of
trial-and-error method for goal attainment).
Skilled performers already know a fair amount about what kinds of
x_r and y_r need to be perceived in order for a given z_r to be
perceived, without having to blunder around in this space. For
example, skilled screwdriver user knows what postures you need to
assume to exert a lot of force in various directions, combinations
of joint angles you wouldn't bother with if you only needed small
rather than large levels of force. So there's a fair amount to say
about what this kind of skill consists of and how it's acquired, but
F&T didn't point the way to saying it, though they ought to have been
able to.
Avery.Andrews@anu.edu.au