[From Bruce Abbott (2000.10.05.1255 EST)]
Bill Powers (2000.10.05.1153 MDT) --
Bruce Abbott (2000.10.05.1025 EST)
The pleasure experienced when drinking the water provides a
universal code signifying immediately to the organism that drinking water is
the right thing to be doing. (It _feels_ right!) The organism will thus
drink even when the error that arouses drinking may not be immediately
corrected by the act. (I am thinking here of the need for the water
consumed to be absorbed in the digestive tract and transported to the body's
cells, all of which takes time. If pleasure results from the reduction of
error, then drinking motivated by this error would produce no immediate
pleasure.)
There is absolutely no function you propose for either pleasure or pain
that would not be fulfilled by a higher level of control system. The
problem here is that you're trying to give pleasure and pain an existence
separate from _what_ is pleasant or painful, the perception itself. This
reification is unnecessary, and anyway you fail to propose any mechanism
that could accomplish it.
By calling it reification, you assert that pleasure and pain are mere labels
and not real perceptual experiences. Pleasure and pain seem very real to
me, I assure you, and I suspect that human beings who have not yet acquired
language, as well as dogs, cats, rats, etc. feel these experiences too
without the benefit of linguistic labeling. If these experiences enter
consciousness, then there must be specialized circuitry in the brain that
produces them, just as the experience of, say, vision, depends on the
operation of specialized neural perceptual analyzers in the visual system.
As to the charge that I have not proposed a mechanism that could accomplish
it, I plead guilty. At this point it's just an idea I'm exploring, and
these discussions are helpful in focusing my thinking about it. Perhaps in
the end you will convince me that it's unworkable, but at this point it
seems promising to me and worth thinking through.
. . . Certain sensory experiences
will evoke, through connections in the brain, feelings of pleasure or
displeasure; usually these will have been associated through evolution or
learning with physical states that in the history of the species or organism
have fostered continued survival. By controlling for pleasure (or
elimination of displeasure), the rat also controls for being properly
nourished, hydrated, groomed, and so on.
Wrong way around. By controlling for being properly nourished, hydrated,
groomed, etc., the rat controls those things it desires -- which it the
same as saying it controls what is pleasant for it. "Pleasant" means no
error; "unpleasant" means error -- and, I claim, nothing else.
Well, correct me if I'm wrong, but I thought that in PCT, errors are not
consciously perceived. But assuming that they are, the question becomes one
of how errors in numerous (but not all) control systems get perceived as
pleasant or unpleasant -- what is doing the perceiving? There must be a
neural circuit which, when moved into a particular state, gives rise to
these conscious perceptions of pleasantness or unpleasantness, just as a
certain state of neurons in the visual cortex produces the experience of,
say, red. Pardon me for saying this, but you haven't proposed any mechanism
that would do this.
Let us assume that such circuitry exists (as experience shows they must).
These conscious feelings along an unpleasant-pleasant dimension must be
aroused by inputs to such a circuit organized in such a way that states of
perceptual input that are bad for the organism's well-being bias the state
of the circuit toward the unpleasant end of the scale, whereas those states
of input that are good for the organism's well-being bias the circuit toward
the pleasant end. One way to define "bad" or "good" is in terms of error in
a perceptual control system, or better yet, both error and change in error.
Another is to define "bad" or "good" as certain perceptual states that
correlate with such error -- such as thirst -- and rate and direction of
change in the intensity of such inputs. One might also define certain
perceptual input states as inherently "good" or "bad," although this
valuation could be overridden by other input factors. For example, certain
foods would taste good even when the body is not in a state of need for the
nutrients they provide, and one eats for the pleasure of the experience, not
because eating thereby corrects some underlying physiological need. This
pleasant experience can be reduced in intensity or even moved into the
unpleasant region by changes in other bodily states, as when one is ill.
Such a mechanism may be complex, but there is no reason to insist that each
system serving a particular biological need could not have its own local
circuitry to yeild relationships such as those just described.
Even as undeveloped as it is, this proposal seems in principle capable of
accounting for certain observations that are more difficult to explain (it
seems to me) via straight PCT. For example, people sometimes starve
themselves during the day in anticipation of a wonderful meal they intend to
consume at dinner time. The self-imposed starvation creates errors in
certain physiological controlled variables such as stomach fullness and
blood-sugar level. The person not only feels hungry, but certain sensory
experiences -- the tastes and smells of the food that will be consumed at
dinner -- will now produce added pleasure. The expectation of this added
pleasure at dinnertime is what motivates the person to skip breakfast and
lunch and thus endure a somewhat unpleasant period of hunger.
This proposal is not intended to replace hierarchical control; rather it is
an elaboration of the hierarchical model. When the person is going without
breakfast and lunch in order to heighten the pleasure experienced at dinner,
this is clearly a control process. When the person is having a malted milk
even though she has recently eaten a full meal and clearly has no bodily
need for the malt, the acts involved in obtaining and consuming the malt are
still the products of the same old control processes, but what is being
controlled for in this case is not the correction of a nutritional error but
rather the pleasure obtained in the act of consumption. That pleasure may
exist because in our species (and many others), sugary foods are generally
good for survival, and so we have evolved an innate linkage between
sweetness (at a certain range of intensities) and pleasure, with the gain of
the system tied to other relevant variables such as hunger (itself tied to
error in nutrient control).
All right, what exactly is your model of that "mechanism?" I say the model
is a control system. To "like" something is to set a nonzero reference
level for it; to dislike it is to set a zero reference level for it. That
_is_ the mechanism; no other is needed.
Vague as it is, I believe that the outline sketched above provides a fair
idea of the sort of mechanism that would be required. I say that if you
like something, then (if other factors don't conflict), you set a nonzero
reference level for it. After that, both views make the same prediction --
the individual will control for experiencing some positive level of the
variable.
I'm afraid that you've misunderstood my proposal. If you go back and reread
it, you will find that I have indeed considered the concept of error signals
in specific different control systems. When error exists in a certain
specific control system, certain specific sensory experiences become
pleasurable (although they may be at least somewhat pleasurable even without
such error -- as witness the eating of sweets after consuming a full meal).
And you have misunderstood mine. I am saying that the concept of pleasure
as a signal different from any in the control systems is superfluous.
You can design a robot without such signals that keeps the variables in
question in control, but it won't always behave like a real animal or
person, for which the pleasure-displeasure dimension of its perceptual
inputs is a crucial determinant of what variables will be controlled, and at
what levels. Unless, of course, you observe what variables are being
controlled first and then design your control-system simulation around that
knowledge.
When you are thirsty and drink, just how much trouble do you have in
determining that it is the water you are drinking that is producing all
those really good feelings? How much trouble do you think a rat would have?
None at all, under my proposal. "All those [hypothetical] good feelings"
are the thirst decreasing (as well as any other errors that may decrease as
a consequence, such as the fear of dying from thirst), and nothing else.
That is _why_ it's so easy. But in saying it is easy, and no trouble, you
do not explain why it is so easy and so little trouble. You say nothing at
all about what the mechanism is.
Again, to my knowledge your proposal has no mechanism to perceive either
error or any pleasure/displeasure that would supposedly arise from it. That
is what Bruce Gregory is trying to say when he notes that the thermostat
takes no pleasure in the fact that the temperature is moving toward its
setpoint. We living creatures do, although strictly speaking (as you
emphasize) the human "thermostats" could all do their jobs perfectly well
without any sort of evaluative mechanism at all -- as they apparently do for
many physiological variables such as blood pressure.
Why should we experience these states at all? Why should we label these
experiences as good or bad, pleasant or unpleasant? What is doing the
evaluating? For what purpose?
The control system is doing the "evaluating" such as it is. But the
evaluation is embodied in the fact that we have nonzero reference levels
for some things and zero reference levels for others. At the same time, our
thinking levels can form cognitive opinions about our experiences which
_reflect_ the reference settings, and can influence them in the future. The
taste of cool clear water when we are thirsty is made of many perceptions,
and because of the way we feel when we drink it we set the reference values
quite high when we are thirsty -- but not when we are not thirsty. Why
should setting the reference level to zero change the associations with the
perception from pleasant to unpleasant? The answer is that it doesn't,
because there are no "associations."
The phrase "because of the way we feel" in the third sentence above contains
the very thing you seek to deny -- the way we feel, in addition to the
wetness and the coolness, is pleasure. Indeed, if you had just stumbled in
from the Sahara after five days without water, the experience might better
be described as ecstasy. I agree with the rest of that sentence: ". . . and
because of the way we feel when we drink it we set the reference values
quite high when we are thirsty -- but not when we are not thirsty." (You
see, these two models are not so far apart.)
My point is that we don't need anything outside the perceptual hierarchy
and the control hierarchy to explain what we mean by pleasant and unpleasant.
My point is, obviously, that we do. Your position is equivalent to stating
that we don't need a perceptual apparatus to account for control of a
perception, we just need to know the values of the CV.
You suggest that pleasure/pain are
classifications based on error, but in your system the classification has no
functional role to play. Apparently it is simply a mental game we play for
our own amusement.
Not at all. If we perceive something as pleasant, we may plan on
experiencing it again, and the categorization thus plays a part in our
control processes.
How is error in a control system consciously perceived? Why aren't all
errors consciously perceived?
I say that the pleasantness or unpleasantness itself is a perceptual
experience, an integral part of the perception aroused by the sensory inputs
themselves, and dependent on the state of error in those control systems for
which those sensory inputs correlate strongly with physiological effects
that close the negative feedback loop. Without this label being attached to
the sensory experience, becoming part of it, we simply have no basis for
choosing among the various alternatives, as Damasio points out so
forcefully.
But that just isn't true. There is _always_ a basis/
Sorry, I thought you had read Damasio ("Descartes' Error"). Damasio notes
that certain brain-injured patients remain as intelligent (as indicated by
standard I.Q. tests and related measures) as before the injury, and when
given a set of facts and a decision to make based on those facts, they can
generate all the relevant scenarios and the consequences of any given
decision. What they cannot do is decide. For them, each possible outcome
seems just an outcome, neither good nor bad, leaving them with no criterion
for deciding which course of action to adopt. As a result, when they do
finally decide, the decision is often disasterous. They simply don't _feel_
one way or another about the options.
A bland bowl of cold mush and a hot, juicy New York Strip may
both correct an error in our nutritional state equally well; given the
choice, we may choose based on which we expect to arouse the most pleasure.
And there is nothing in this proposal demanding, as you claim, that pain or
pleasure exist as some disembodied entity unassociated with specific sensory
experiences, or that different systems can't be in different states.
But why do you confine your reference signals to "nutritional state?" How
about the taste itself? If you have come _for any reason_ to set a high
reference level for experiencing some taste, you will seek that taste
regardless of the nutritional effects of consuming whatever gives that
taste. In other words, you will act and feel in all respects as if the
taste is pleasurable. At a higher level, you may even have a positive
reference level for the idea that you are here having this sought-after
experience -- for example, you might get snobbish satisfaction out of
eating caviar even though at a lower level the desire for experiencing the
taste of caviar can be satisfied with a vanishingly small portion of it.
You're sounding very Jamesian here -- I see the bear, I run, I perceive
myself to be running, I conclude that I am afraid. Or here, I find myself
moving to the pantry, opening a bag of potato chips, and eating them,
therefore, I conclude that I like potato chips. Maybe that works for you.
It doesn't work for me.
I think we're having an Occam's-Razor kind of argument. You want pleasure
and displeasure to be a separate phenomenon, while I am claiming that there
is no need for this to be the case. If we just drop the notion of a generic
kind of experience of pleasure and displeasure, and let the state of the
control systems in the brain substitute for them, we will lose nothing at
all in explanatory power, and we will shed two constructs that are not
needed. The _very same experiences_ to which you refer will still occur;
only the explanation will be different.
I think we agree one should simplify as much as possible to account for the
phenomena to be explained, and no further. Where we seem to disagree is
whether the PCT account as currently developed fails to capture certain
phenomena of human experience. I have suggested that it does so fail,
because it does not include certain design features that are needed to
handle such phenomena, and have proposed in a general way what additions
might be needed.
Bruce A.