[Avery Andrews 931106.2016]
Well, here's how I see the feedforward/feedback issues raised by Hans
Blom. In the case of walking in a darkened room, I think there are
two possible ways of looking at it:
a) precompute actions that have a desired effect, and perform them
(using kinesthetic feedback to control the performance)
b) feedback as usual, but with a more complex and indirect perceptual
function being used to provide information about where you are
in the room.
Hans seems to assume (a), but the (b) interpretation, which I take to
be Tom Bourbon's, follows naturally from the idea that feedback systems
don't directly control some objective variable that an external analyst
might deem to be ecologically significant, but just some perception,
which can be an arbitrarily complex (and time-dependent) function of
the organism-environment system. (However, due to evolutionary and
reorganizational selective effects, the perceptions do tend to correlate
pretty closely with environmental variables that look significant to
external analysts.
But organisms always control some approximation to
the `best' variable, because the cost of perceiving a better one would
exceed the benefits, in the environment they have evolved in: if
birds had sonar as well as vision, they wouldn't crash into glass doors,
but this would be to energetically expensive to actually be worth it.
Sometimes the approximations are rather cheap and nasty, as when
frogs lunge at crows in the middle distance, as well as flies nearby).
On the (b) interpretation of walking in a darkened room, the general idea
is that people (and presumably other mammals, and perhaps even other
organisms) are able to use multiple sources of information as input
to perceptions of such things as where they are, the presence of
predators, etc. Normally, vision is the best source of evidence about
where you are, and in the presence of light, the backup systems make
no discernable contribution. Switch off the light, and a perception
of where you are continues to be maintained on the basis of such things
as double integration of signals from the vestibular canals, plus
kinesthesia, maybe even efferent copy (if the efferent signals copied
are of reference levels for lower level control systems, they can be
assumed to represent what is actually happening). I have heard the
word `imagination' used in this context, but I think this is a mistake:
we're not manipulating imagined perceptions of where we are, but an
actual one, on the basis of things that are actually happening. I am,
however, eliminating any in-principle distinction between a perception
and a model, except maybe that a model might be regarded as a perception
based on multiple sources of sensory information.
So then we come to the question of whether (b) is actually better than
(a), or is just a way to make some embarassing observations look
consistent with PCT without actually saying anything useful about them.
In this particular case, I think (b) really is better, since it explains
more about what goes on under various circumstances. If I bump into a
chair while moving the darkened room, I don't have to continue
following my program, and plough through it, but simply move around it,
as a CROWD person might (it's my impression that under these
circumstances, you know what direction you want to be going in, but
are fuzzier about distances). On the other hand if I encounter a wall,
its much more upsetting, because my perceptual world falls= apart and I
become disoriented, no longer thinking that I know where I am.
Furthermore, the problem of switching from `feedback' to `feedforward'
mode gets somewhat encapsulated, and hopefully simplified, taking the form
of the problem of maintaining perceptions at useful values when some
the contributing information becomes unavailable.
The steering-on-an-icy-road incident reveals another subtlety of
control, (perhaps limited to apes?), which is control of expected
consequences. Presumably Hans knew that yanking the steering wheel
hard would (a) not solve his problem if the no-traction circumstance
persisted (b) cause him to loose control if traction were suddenly
regained. A less well-trained and/or cool-headed person would have
done it anyway, but Hans took an option that gave him a chance, over one
that gave him little or none. In humans, control of imagined
consequences is to a large degree achieved by following culturally
transmitted rules of thumb. E.g. if you don't sweep your floors
regularly, undesired life-forms will take up residence in your house,
a rule which `civilized' girls learn from their mothers.
When indigenous peoples take up aspects of the Western life-style,
they tend not to immediately pick up the rules that go with
it, with nasty results. Other such rules are presumably picked somehow
from experience: when chimpanzees avoid looking in the direction of
food they know about in the presence of their superiors, it's presumably
because they've noticed that when they look in the direction of the food,
the superior goes there and gets it, which is something they don't what
to happen.
On the other hand, the ability to anticipate and avoid undesired
results in unfamiliar circumstances is not very widespread - an early
literary example in our tradition would be where Odysseus
refrained from stabbing the Cyclops in his sleep, because he realized
that with the Cyclops dead, the Greeks wouldn't be able to shift the
stone from the door of the cave, and they'd all die. Presumably
this got into the Odyssey because it was seen as a significant
manifestation of intelligence, and something which many actual people
would in fact fail to foresee, at least under stressful circumstances.
So again we have a choice: are we following pattern-action rules
(when you see dirt on the floor, sweep it up), or controlling for
predicted consequences (controlling for a perception `this will/will not
happen', or `this is/is not likely to happen'? It may well be that
both of these things happen: the fact that people have many customs
and traditions that they can't give coherent reasons for suggests to
me that mindless pattern-action rules do have a certain role to play.
(`Salute when you seen an officer' might be one that gets drilled into
people in the military, though this is not something I know anything about).
The difference may be hard to discern, because much of the reasoning
whereby consequences are predicted also takes the form of pattern-action
rules, but the pattern-action-rules should be able to be made to
betray their presence by a certain rigidity and a tendency to be
followed even when they are clearly serving no useful purpose.
(I assume, of course, that the `actions' in these rules are specified
as perceptions).
Control of predicted consequences on the other hand clearly involved
when people choose a course of action by actually imagining
alternatives, trying to figure out where they think they will lead,
and chosing the one that seems best.
In Hans' particular case, we don't know whether he (a) figured out
from first principles that yanking the wheel would be a bad idea
(b) had gotten some serious training and actually been taught what
to do under these circumstances (c) read something somewhere, and
managed to apply the knowledge. I would see (a) or (c) as genuine
control of predicted consequences (controlling for `I will be less likely
to loose control of the car'), while (b) might involve a significant
degree of pattern-action rule.
Well, that's all I have time for now. I will probably be pretty
inactive for the next few months - lots of travelling coming up.
Avery.Andrews@anu.edu.au