Anticipation, high level stuff

[Avery Andrews 980106 Eastern Oz]
  (Bill Powers 980105.1605 MST)

I fully agree with the general point of the post, but on the other hand
these questions about anticipation just keep coming up again and again
and it seems to me that there really need to be some sensible and
coherently described possibilities to present to people who `drop in'
on CSGNET.

So here are some possibilities for `anticipation'. One is what I might
call `model assisted perception'. These are perceptual functions that
use some kind of predictive scheme (a partial world-model, as tiny
as possible) to update a perceptual signal when the preferred form of
evidence is not available. For example in keeping track of moving
objects, if you record position and velocity observation about an
object, you can use the velocity info to continually update the
position until you can look that way again, and check and reset
everything. The neurologist Oliver Sacks in one of his books describes
a patient whose motion-detectors got destroyed; one of her problems
was that she couldn't stand to be around moving people and objects,
since she'd look around, and everything seemed still, but then she'd
look again, and things and people would be in new, totally unexpected
places. She found parties very disturbing, and traffic extremely
dangerous. This suggests to me that humans might actually employ a
system of this kind (for example to help with collision avoidance).
Something like this is also employed by the Quakeworld game server, that
makes internet action-gaming possible in spite of lag.

One might call this anticipation (looking again where a previously
seen object is expected to be, moving aside from where another
approaching person is expected to be), but it's really quite within
standard PCT (whether or not it actually exists in any particular
organism), because it's *just* a perceptual function, regardless
of whatever verbiage might be employed in describing it.

Another possible form of anticipation is control of perceptions that
can be described functionally, from an observers point of view, in
anticipitory terms. Imagine a perceptual function that evaluates
the visible surroundings, noting presence of people and precipices,
the distances between them, the age, activity level and state of
sobriety of the people, etc., and produces a number that corresponds
roughly to the probability that someone will plummet to their death
within the next five seconds. A control system that tried to maintain
this perception at a reference-level of zero might be described
as `anticipating the likelihood that someone will plunge to their
death, and trying to prevent it', but that's just a functional
description of what it does, useful for some purposes but not very
indicative of what it is. Again, it's just another perceptual circuit
(I think parents have lots of systems of these kinds w.r.t their
children; it's notorious that they're not very good at making up
verbal stories about why they have forbidden various things, indicating
that they're not relying on reasoned extrapolations of possible
disaster from what the kids are wanting to do, but on something else).

I think an extreme case of this sort of thing arises in language,
with grammar; I take grammar to be a system whereby speakers
attribute changes in state of mind to their audiences, on the
basis of what the speaker has perceived the audience to have
heard (or seen, depending on linguistic modality). So if I say to
my mother `I arrive at Philadelphia airport at 3:15PM June 18',
I thereby perceive her to know this (unless something else happens
to revise this perception; `defeasability' is an important feature
of the higher-level perceptions). Here we're using a pretty complicted
model to attribute to other people states (of mind/knowledge) which
we think of as existing at the present time (rather than the future),
but which aren't directly observable at the present time, but do
tend to manifest themselves as dispositions (if my mother doesn't get
totally furious with me, and if nothing goes wrong with her car or
her schedule, I can expect her at the airport at the time I said
I would arrive (actually, I can't imagine doing anything that would
upset her enough so that she wouldn't try to collect me at the airport)).

Why speculate about this kind of stuff? why not just do bugs first, and
go on to other stuff when they're sorted out? I certainly agree with
doing bugs, but if passing spectators are interested in humans, language
& other `higher level' stuff rather than bugs, there's got to be some
indication of how one could envision treating these things, in order to
show that they're not beyond the scope of the model. Keeping in mind
that modern academics are (and pretty well have to be, in order to keep
their jobs) surpassing experts at making up reasons why they shouldn't
take inconvenient things seriously, so that if you give them the
slightest excuse to dismiss something, they'll take it.

And one way of taking some of the appeal out of terms such as
`anticipation' is to suggest a number of different mechanisms whereby
a functional effect describable from the outside as `anticipation'
can be achieved, so that people will be encouraged to see this term
and its mates as vague and useless.

Avery.Andrews@anu.edu.au

[From Bill Powers (980105.2128 MST)]

Avery Andrews 980106 Eastern Oz --

I fully agree with the general point of the post, but on the other hand
these questions about anticipation just keep coming up again and again
and it seems to me that there really need to be some sensible and
coherently described possibilities to present to people who `drop in'
on CSGNET.

I like all your thoughts on this subject. Your proposals about how
anticipation might be accounted for are all plausible. The problem is that
there are so many ways to account for specific instances that could be
called anticipation that I wonder why anyone thinks it's really a problem.

Here's another. Take the case where a child finally learns to look at the
place where a toy will reappear after going behind a visual obstruction.
The sequence is "disappear here, reappear there." I believe that when
children learn this, they don't "track" the imagined object as it moves;
they transfer their gaze to the expected reappearance point and simply
wait. They expect the sequence to be completed. So, in a just-so-story way,
I can see how sequence control could account for this phenomenon.

Here's another, slightly more complicated. Suppose you're holding out an
arm and pointing a finger at an object traveling at a constant speed, that
disppears for a short time behind another object. You continue swinging
your arm at a constant speed, and when the object reappears your finger is
nearly on it, requiring only a brief correction. So work backward. If the
kinesthetic position control system is moving the arm at a constant speed,
the position reference signal is changing at a constant rate. If it is the
integral of a kinesthetic velocity error signal, there is a constant
velocity error signal. If the velocity reference signal is constant, and is
the integral of a relative-position error signal, then the
relative-position error signal is zero (the integrated output is neither
increasing nor decreasing). The relative-position error signal can be zero
if the relative position matches the reference signal, OR IF THE REFERENCE
SIGNAL IS ZERO, TURNING THAT SYSTEM OFF. So to keep the arm "tracking" the
invisible target, all that is necessary is to turn off the
relative-position control system in such a way that its output remains
constant (the integrator doesn't decay to zero). This accounts for the
continued apparent tracking, without bringing in anything that
"anticipates" or "predicts" or even "imagines."

This is another just-so story, of course, because I haven't actually
simulated this kind of system. When somebody gets around to actually
working out a model for comparison with how people actually do this,
something like this will probably be found even if I haven't got it quite
right. I'm sure enough that this effect can be created with ordinary
control systems, so it doesn't worry me.

Since I can't get any closer than that, I sort of tend to lose interest; if
I can think of one simple answer, that probably means there is a simple
answer even if it isn't mine.

I guess what I don't understand is what kind of answer anyone _expects_.
Precognition? Some mysterious ability to react to things before they happen?
Quantum tunneling? When people nag at me to explain this and explain that,
I tend to get impatient because they don't seem to realize that it's just a
matter of circuitry. If you can explain to me just what the system is
supposed to do, I can probably come up with a circuit that will do it, or
two or three circuits including but not limited to control systems. After
all, besides working on PCT for 45 years, I also made a living designing
electronic circuits, both analog and digital, to do a lot of different
things. Circuitry is not a problem compared with the problem of defining in
some useful way what the circuit has to do.

I think that anticipation is a vague term that refers to a lot of different
things that have little to do with each other. It can mean anything from
adding 10% first derivative to a proportional perceptual signal to driving
to the airport expecting to see a familiar face at the gate. Surely these
processes are not so similar that they can legitimately be described by the
same word!

What makes all this even more frustrating is that I don't have any idea how
the BRAIN does these things. I might know how I would do some of them if
given the specific problem to solve, but since there is always a large
number of circuits that will perform the same function, I have no way of
knowing whether my guess is anything like what the brain actually does. And
of course some (many!) things that the brain does are so complex that I
can't think of even one circuit that would do the same thing -- so how can
I guess how the brain does it? How does the brain recognize a spinning
cube? All I can be sure of is that it doesn't happen by magic.

I really hate to give blow-off answers to questions like these, but if one
doesn't do that, if people won't accept "I don't know" as an answer, what
else can one do but make up some plausible story just to turn the noise off?
And what can you do when you see other people who can't stand saying "I
don't know" making up their own fairy tales and passing them off as the
latest hot poop? It's awfully tempting to join that game -- to fight
bullshit with bullshit. But there's no end to it -- as soon as you think up
one good story, there's someone else coming back with "Yeah? well then how
about _this_?" It's like playing tennis against one of those automatic
machines. As soon as you hit one ball back, there's another coming at you.
You can get so busy hitting balls back that you forget this isn't a real
tennis game.

And one way of taking some of the appeal out of terms such as
`anticipation' is to suggest a number of different mechanisms whereby
a functional effect describable from the outside as `anticipation'
can be achieved, so that people will be encouraged to see this term
and its mates as vague and useless.

Right on the mark, but you're still not playing tennis.

Best,

Bill P.

[Martin Taylor 980109 00:31]

Bill Powers (980105.2128 MST) to Avery Andrews 980106 Eastern Oz --

I fully agree with the general point of the post, but on the other hand
these questions about anticipation just keep coming up again and again
and it seems to me that there really need to be some sensible and
coherently described possibilities to present to people who `drop in'
on CSGNET.

I like all your thoughts on this subject. Your proposals about how
anticipation might be accounted for are all plausible. The problem is that
there are so many ways to account for specific instances that could be
called anticipation that I wonder why anyone thinks it's really a problem.
...
I really hate to give blow-off answers to questions like these, but if one
doesn't do that, if people won't accept "I don't know" as an answer, what
else can one do but make up some plausible story just to turn the noise off?

I think it's fine to give "just-so" answers in the situation Avery
postulated. The usual context is a question "Can your theory cover that"
(with the obvious expectation that the question will be answered: "Well,
not yet, but we expect it to" or "No, we never thought about that problem.")
A perfectly fine answer is: "Yes, there are all these different ways, and
probably a dozen more, but nobody has done the experiments that would
let us choose among them. Do you want to?"

I remember my first few month on CSGnet. Many a time I would envisage a
situation or an observation that I could not see a way to handle in PCT.
Such an answer would have satisfied me--and quite often, that's the kind
of answer I got. (Whether or not I took up the challenge is a different
question:-)

So, I wouldn't consider your answers "blow-off." You see them that way
because you are so familiar with PCT, and it's obvious to you that there
are lots of ways it might happen--whatever "it" may be. How "it" might
happen is not obvious to the new or casual "attendee" on CSGnet. It may
pain you to give an answer when "I don't know" would be more exact, but your
"blow-off" answer usually would adequately answer the implicit question.

Martin