Rick Marken writes:
What are "internal dynamics"? If, by this, you mean "a model of the
nervous system" then I'd say that is certainly a way to describe the
problem as I see it as well . . .
By "internal dynamics" I simply mean the time-dependent input/output
properties of the animal. If you are interested in understanding the
mechanisms underlying natural animal behavior, then this might involve
modeling the nervous system. However, it needn't necessarily do so.
If you are designing an artificial agent, the necessary internal
dynamics could be implemented with ping-pong balls and peanut butter
as long as the external behavior of the agent were appropriately
coupled to its environment.
Would the acceleration of the bug as it falls off a ledge count as a
behavior to be modeled? If not, why not. If so, why so?
It certainly could. Some species of moths are preyed upon by bats
who, as we all know, navigate by echolocation. These moths have
evolved an interesting escape mechanism. Whenever they detect
vibrations of a certain frequency (namely, that used by bats searching
for prey), they simply fold up their wings and drop like a stone.
This certainly counts as a behavior in the ethological sense, though I
couldn't say whether it is a behavior in the specialized technical
sense of PCT. By the way, what could possibly constitute the
controlled variable in this case?
Both Rick Marken and Bill Powers summarized the basic tenets of PCT
and HCT for me (which I will refer to as simply CT for simplicity). I
will try to respond to their summary below. If some of my comments
are based upon a misunderstanding of CT, I trust that one or both of
them will correct me.
The basic idea of CT seems to be that behavior is the consequence of
negative feedback control of selected sensory inputs. The use of
negative feedback control to regulate important variables is clearly
ubiquitous in biological systems. If this is the only point of CT,
then I can't really find any reason to disagree.
However, if CT is making the much stronger claim that negative
feedback control is universal and ALL behavior can be understood in
its terms, then I am extremely skeptical. Animals certainly DO
control some sensory inputs using negative feedback, but I don't
believe that they JUST control sensory inputs. Many consequences that
are of utmost importance to an animal are not controlled in any
negative feedback way.
To take just one example, the American cockroach has an escape
response: Whenever anything lunges toward it (i.e. a striking toad or
a foot), it turns roughly 180 degrees away from the direction of the
attack and runs away. A fair amount is known about the neural
circuity underlying this response (we have actually been involved in
modeling this system for a couple of years) and people have found that
it is organized not as a negative feedback system, but as a
feedforward system. Upon reflection, the reason for this is rather
obvious. The sole purpose of the escape response is to make sure that
the cockroach isn't where it was when the attack began. Consequently,
the cockroach can complete its initial turn in about 60 MSEC. Given
the latencies involved in the sensory organs and neural signal
transmission, there simply isn't time to do negative feedback control
of this turn. Nor is such precision necessary.
But wait, the story gets even more interesting. It turns out that the
cockroach can factor a great deal of contextual information into its
escape, including auditory, tactile, visual, and proprioceptive cues.
Again, the necessity of this is obvious upon reflection. An attack
can come at any time: when the insect is in midstride, when it is
feeding, when it is near the edge of a wall, etc. The actual
movements required to escape might be very different in each of these
cases. For example, experiments have shown that cockroaches that are
attacked near a wall will make entirely different movements than
free-ranging cockroaches. It turns out that there is a population of
about 100 interneurons whose job it appears to be to integrate all the
appropriate contextual information and essentially always be ready to
generate an appropriate escape should the insect be attacked in the
next instant.
And some important variables aren't likely to be controlled at all,
even in a feedforward way. Survivability is undoubtedly one of the
crucial variables for any animal. But do you seriously believe that
an animal explicitly estimates its current survivability and that some
high level control system actually uses the error between this
estimate and a "reference" level to guide its behavior? Such
variables are simply too complex to explicitly estimate, nor is it
necessary to do so. At best, animals may control variables that are
sufficiently correlated with survivability that, on average, they do
in fact survive for an extended period of time. This also suggests
that many variables that appear to be explicitly controlled on casual
inspection may, in fact, not be.
What I was trying to say in the summary of my own position was that
evolution selects for animals that always generate the appropriate
consequences PERIOD. Sometimes those consequences may be generated by
negative feedback control. Sometimes those consequences may be
generated by feedforward circuits (another example of this would be
central pattern generators). Sometimes those consequences may be
generated by the laws of physics (e.g. the plummeting moths mentioned
above). Evolution selects only for the viability of the complete
package and for that reason I think that it can be grossly misleading
to impose our organizational preconceptions on evolved control
systems.
Best regards,
Randy Beer