Variability in Brain Function and Behavior
Paul Grobstein
Department of Biology
Bryn Mawr College
Bryn Mawr, Pennsylvania 19010
(610) 526-5098 pgrobste@brynmawr.edu
published in The Encyclopedia of Human Behavior, Volume 4
(V.S. Ramachandran, editor), Academic Press, 1994 (pp 447-458)
copyright by Academic Press, Inc.
made available on Serendip
I. The Problem of Variability and its Significance
II. Behavior and the Nervous System: Input/output Boxes Within Input/Output
Boxes
III. Intrinsic Variability in Input/Output Boxes at All Levels of
Organization
IV. The Nature of Intrinsic Variability
V. The Analysis and Significance of Intrinsic Variability
VI. Conclusions
GLOSSARY
Information. An increasingly difficult and important term to define.
Central to the concept is that there are aspects of reality which need to be
characterized not in terms of matter or energy but rather in terms of the
particular organized forms taken by one or the other or both. Similar
information, for example, can be embodied as signals in motions of the air
(sound waves), ink on a page (printed text), or a series of ones and zeros (on
a hard disk in a computer).
Input/output device for information Any physical system whose primary
function is to receive, transform, and emits signals (information, as opposed
to matter or energy, see Information). Examples include thermostats, radios,
computers, neurons, and the brain.
Levels of organization A term emphasizing the reality that virtually
all interesting systems (the brain and behavior providing two examples) are
made up of identifiable subsystems, which are in turn made up of identifiable
subsystems, and so forth. As a general rule, the properties displayed at one
level of organization reflect the properties of the subsystems at the next
lower level of organization together with the particular way those subsystems
are organized in relation to one another. The properties of the subsystems
typically constrain but do not determine this organization (which reflects
information over and above that inherent in the subsystems). For this reason,
interesting system can productively be explored at all levels of organization,
and need to be to be fully understood.
Membrane potential. The electric field which exists across the limiting
membrane of a neuron (and many other cells). The value of the potential at any
given time reflects the current and recent movement of charges across the
membrane due both to intrinsic properties of the neuron and signals received
from other cells. The value of the potential at any given time in turn
determines the signals sent to other cells.
Nervous system The most sophisticated known input/output device for
information. The term is a general one which, in animals with backbones
(humans and frogs among them), includes the brain, spinal cord, and other
related components. Other animals (including leeches) have nervous systems
which, while somewhat differently organized, are fundamentally similar in terms
of their basic function.
According to the Harvard Law of Animal Behavior, "under carefully
controlled experimental circumstances, an animal will behave as it damned well
pleases." An informally propagated and often ironically intended summary of
large numbers of observations, the Harvard Law in fact has quite concrete and
deep significance for understanding the basic information processing
characteristics which underlie the behavior of all organisms, humans very much
included. An appreciation of this requires drawing together threads from a
variety of lines of inquiry, and is facilitated by a perspective which treats
both behavior and the nervous system as nested sets of interacting
information-processing boxes each with more or less clearly defined inputs and
outputs. Briefly put, the Harvard Law provides the basis for a desireable and
productive fusion of scientific and folk perspectives on the determinants of
behavior, one which acknowledges that some degree of unpredictability is not
only inevitable but desireable.
I. The problem of variability and its significance
A fundamental question about behavior, whether explored informally as all
humans do all the time or formally as done under experimental conditions by
scientists, is why a given organism at a given time under given environmental
conditions behaves in a particular way. For many investigators,
"understanding" behavior is equated with being able to predict for a given
subject what behavior will occur at a given time. For others, the ability to
study particular components of behavior depends on being able to reliably
observe that behavior. In both cases, many investigators are predisposed to
treat behavioral unpredictability as not only an inconvenience but, quite
commonly, as a clear indication of the failure of the experimenter to control
the circumstances with sufficient precision so that meaningful observations can
be made.
This general perspective on variability has contributed importantly to a
style of inquiry into behavior whose record of successes is both long and
enviable. At the same time, it contributes to a dissociation, to the detriment
of both, between the more formal and "scientific" study of behavior, and the
common sense informal body of wisdom people are constantly creating in their
day to day lives ("folk psychology"). Explanatory frameworks and models of
behavior emerging from mindsets predisposed to dismiss variability tend, for
obvious reasons, to have a deterministic character to them. These impact less
than they might on folk psychology both because they appear somewhat
threatening to human dignity, and because they fail to speak to a question
central to informal understandings of behavior: they explain why a behavior
occurs when it does, but not why it doesn't always occur. Conversely,
the very real insights from the common sense body of wisdom about behavior,
phrased as they are in terms of both puzzlement about its unpredictability and
enjoyment of its playfulness, impact less than they might on more formal
inquiries.
Mindsets, of course, influence the accumulation of data sets, and
there is a corresponding bias againt the reporting of variable data
sets in the scientific literature. This of course in turn influences
mindsets, reinforcing those that dismiss variability as a meaningful
and explorable phenomenon. Against this background, the existence of
descriptions of behavioral variability in the biobehavioral literature
is quite significant, and such descriptions do indeed exist. In the
area of movement control, for example, there is a well-established
body of phenomena under the rubric of "motor equivalence": under
apparently identical task descriptions and circumstances, the
particular movements made by even highly trained subjects vary for
largely unknown reasons from occasion to occasion. Ambiguous figures,
such as that shown in Figure
1, reveal a comparable largely mysterious variability in
perceptual processing: while the particular figure seen can be
influenced by prior suggestion, either form of the figure may be seen
on successive presentations under apparently identical circumstances.
Other examples where substantial variability has been documented
include the processes of language acquisition in humans and song
acquisition in birds. Both involve "babbling" phases during which
largely unpredictable sound sequences are produced. More generally,
the careful observational studies of ethologists have led some to
suggest that the concept of a "fixed action pattern" should be
replaced with that of a "modal action pattern" in recognition of the
reality that even relatively stereotyped behavioral acts are not in
fact identical when looked at more carefully.
In short, the scientific literature does in fact include fragmented
supporting evidence for the Harvard Law of Animal Behavior. Despite this,
there remains in many scientists' minds a general skepticism of the Law, and an
associated reluctance to take it seriously. This is partly a methodological
matter (how is one to study a mysterious variability?), and partly a pragmatic
bet on the odds (if I work very hard to understand this variability it is just
going to turn out to reflect my experimental inadequacies and disappear;
besides, some imprecision and "noisiness" is inevitable in any real system).
It also, however, reflects a deep-seated discomfort (not restricted to
scientists) with the mysterious and uncontrollable, a combination of
characteristics which smacks a bit too much of the supernatural. In the
following sections are outlined a perspective which removes the taint of the
mysterious and supernatural from the Harvard Law (without removing its core or
its interest), providing as well both a methodological framework for further
inquiry and a rationale for a pragmatic judgement that the phenomena are indeed
worth exploration.
II. Behavior and the nervous system: Input/output boxes within input/output
boxes.
Two general presumptions accepted to one degree or another by most
neurobiologists underlie this discussion. The first is that behavior and the
nervous system are in some sense the same thing: all aspects of behavior
correspond in one way or another to phenomena of nervous system structure and
function. The second general presumption is that an understanding of nervous
system structure and function provides special advantages for better
understanding behavior. An implication of the first general principle is that
the Harvard Law ought to be understandable in terms of properties of the
nervous system, and of the second is that knowing something of how the nervous
system works should help to make sense of the Law. In this section is provided
a description of some basic organizational features of nervous system structure
and function relevant to these considerations.
To a good first approximation, the nervous system can be regarded as an
input/output device for information which is made up of interconnected
input/output information devices which are in turn themselves made up of
input/ouput information devices. A convenient stopping point for what might,
perhaps correctly, be regarded as an infinite regress is at the level of
neurons, the cells which represent similar building blocks for all parts of the
nervous system. Each neuron is itself an input/output device, receiving
information (typically from other neurons) on receptor surfaces and
transmitting information (typically to other neurons) from specialized effector
regions. Two aspects of this picture are particularly important to appreciate,
both at the level of the neuron and at the levels of each of the successive
larger boxes of which they are a part. The first is that one can, with a
fairly high degree of rigor, identify and independently monitor inputs and
outputs. The second is that the input/output relationship of neurons is in
general neither obvious nor stereotyped. It depends on intrinsic properties of
the neuron itself, properties which can and do vary from neuron to neuron.
That one can identify and independently monitor inputs and outputs, and in many
cases control the former, provides the experimental ability to show the
dependence of input/output relationships on intrinsic properties and, if one
chooses, to characterize aspects of those properties, a point of substantial
importance to what follows.
The same two principles hold for the nervous system as a whole, the highest
level input/output box under consideration here (there is every reason to
believe that a similar analysis could be usefully extended to still higher
levels of organization, one in which individual organisms represent
interconnected input/output boxes from which emerge, for example, properties of
social organization). Like neurons, the nervous system has reasonably
well-defined inputs (sensory neurons, the generally quite small percentage of
the total neuronal population which consists of neurons whose receptor surfaces
are outside the nervous system, rather than in contact with other neurons) and
outputs (a second distinct set of neurons (including motoneurons), again
generally a quite small percentage, whose effector regions are outside the
nervous system, rather than in contact with other neurons). Here too, the
input/output relations are neither obvious nor stereotyped (individual
organisms behave differently from one another). They depend on intrinsic
properties of the nervous system as a whole, a reality which can (as with
neurons) be both verified and explored because of the ability to control inputs
and monitor outputs.
Between the level of the nervous system as a whole, and that of neurons,
there are a variety of levels of organization at which the description of
interconnected input/output boxes each with significant intrinsic properties
continues to hold. The neocortex, for example, has well defined input and
output connections (restricted to other areas of the nervous system), and an
internal organization which influences its input/output characteristics. So
too does the spinal cord, which is interconnected by input and output pathways
to other parts of the nervous system, and has, as well, associated sensory and
motor neurons. These in turn are made up of smaller interconnected
input/output boxes, each with characterizable and distinct intrinsic
properties, such as different cortical regions and spinal cord segments
associated respectively with forelimbs and hindlimbs. There is no unique way
to describe the interconnected boxes within boxes which represent levels of
organization between neurons and the nervous system as a whole, and it remains
to be seen whether there is some best way. Regardless, the more general point
holds. One can identify (perhaps inevitably with some degree of arbitrariness)
at several levels of organization systems of related neurons with well-defined
inputs and outputs, each having intrinsic organizations which are significant
for determining their input/output characteristics.
III. Intrinsic variability in input/output boxes at all levels of
organization
The discussion of basic features of nervous system organization in the
previous section provides the basis for a useful rephrasing of the Harvard Law
of Animal Behavior in terms which make it not only admissable but productive
for scientific inquiry. What the Harvard Law in essence asserts is that there
is some intrinsic variability in the intrinsic properties of the nervous system
as a whole which influences its input/output relationships. To eliminate the
taint of the supernatural, this property of the nervous system as a whole
should be understandable in terms of the properties of the smaller boxes which
make it up. To make the inquiry productive, it should be both explorable in
terms of smaller boxes and meaningful for the understanding of the largest box
(behavior). Central to all of this is the ability to specify rigorously what is
meant by "intrinsic" variability, and to establish its existence.
The general point is both demonstrable and significant at the level of
individual neurons. The output of a neuron is in general due to a time-varying
membrane potential which most typically results in a time-varying pattern of
action potentials and neurotransmitter release. Given that the inputs to a
neuron are known, it is possible in principle (and, in many cases, in practice)
to ask whether a time varying output pattern persists in the absence of any
inputs (or at least of any time varying signal on the inputs). For many
neurons, perhaps most, the answer is yes. Variations in membrane potential can
persist in individual neurons totally isolated from any input pathways. Under
normal circumstances, the outputs of these neurons reflect not only their
inputs but also the membrane potential variations intrinsic to the neuron. To
put the matter slightly differently, the Harvard Law of Animal Behavior holds
for neurons. No matter how rigorously an experimenter ignorant of intrinsic
variability controls the inputs to a neuron, its outputs for a given input can
and, in many cases will, vary from trial to trial. While this may at first
glance seem mysterious, the underlying cellular and molecular mechanisms are
extensively studied and reasonably well understood. Suffice it here to say
that within neurobiology it raises no eyebrows whatsoever to assert that the
outputs of a given neuron reflect not only its inputs but also variations in
signal processing characteristics which occur for reasons largely or entirely
intrinsic to the neuron.
The ability to eliminate inputs and monitor outputs makes possible, at the
level of the nervous system as a whole, a comparable demonstration and
conclusion that one is dealing not with a simple input/output box, one in which
a detailed knowledge of inputs suffices to predict outputs, but rather with an
input/output box with intrinsic characteristics that vary with time. Figure 2A
shows a continuous record of activity from the output nerves of the isolated
nervous system of the leech, a preparation in which all input paths have been
interrupted. It is clear that variations in output occur in the absence of any
variations in input (and, indeed, in the absence of most if not all input of
any kind). Preparations of this kind can in principle provide answers to the
question that inevitably occurs to most students studying the brain: would a
brain in a dish continue to think? The answer, at least in the case of the
leech, is clearly yes. While much of the output activity in the isolated leech
nervous system has unknown behavioral significance, some periods of this
activity are clearly identifiable by their patterns as corresponding to
behaviorally meaningful behavior. The rhythmic activity following the arrow in
Figure 2A, for example, is the output pattern corresponding to swimming
movements. Though comparablely controlled observations have not, for obvious
reasons, been done with human brains, there is every neurobiological reason to
believe that they too would exhibit variable and behaviorally meaningful output
patterns despite isolation from inputs. In addition, "perceptual isolation"
studies in humans clearly show that variations in brain states not only persist
but may in fact be enhanced (yielding, for example, hallucinations) by reducing
variations in input signals. That people frequently seek out quiet spots in
which to think is unlikely to be a coincidence.
Figure 2B shows that the intrinsic variability present in the leech nervous
system is relevant for understanding the Harvard Law of Animal Behavior. A
cutaneous stimulus delivered to an intact leech sometimes, but not always,
triggers swimming behavior. Such input/output variability might, in principle,
be attributed to slight variations in the cutaneous stimulus itself, or to
differences in other sensory inputs at the time the stimulus is delivered. In
fact, as shown in Figure 2B, input/output variablity persists in the isolated
nervous system. Identical electrical stimuli delivered to a peripheral nerve
sometimes, but not always, triggers swimming. Notice also that, even when the
stimulus does trigger swimming, there is substantial variation from trial to
trial in the latency, swim robustness, and motor activity preceding the
swimming episode. In short, the Havard Law of Animal Behavior holds for the
nervous system as a whole.
Comparable observations have shown that intrinsic variability may also be a
characteristic of the intermediate size boxes of which the nervous system as a
whole is made. The significance of this will be discussed further below. Here
it is only necessary to stress that neither the intrinsic variablity of the
nervous system as a whole, nor that in any of the various intermediate boxes
which make it up, require any appeal to the supernatural. Given intrinsic
variability in neurons, there is every reason to expect its existence in the
assemblies of neurons which make up both intermediate size boxes and the
nervous system as a whole. In short, the failure of an experimenter to be able
to predict the behavior of a given organism at a given time under particular
environmental circumstances does not necessarily indicate either a failure of
the experimenter to adequately control the circumstances nor an external deus
ex machina. It may simply indicate a phenomenon which can be rigorously
documented, the existence of intrinsic variation, and which can, as discussed
further below, be taken as a legitimate subject of study in its own right.
IV. The nature of intrinsic variability
In the following, it will be useful to focus on the "nature" of
variability, the particular form variability takes in a given situation, rather
than on its causes. However, one subtlety in the definition of intrinsic
variablity where issues of nature and cause are unavoidably intertwined needs
to be dealt with first. As is evident from the various phenemona of learning
and memory, the state of the nervous system at a given time depends not only on
its inputs at that time but on the history of its past inputs as well. Such
dependence can certainly contribute to difficulties in predicting the response
of a given organism to a given input at a given time, but it is in principle at
least a correctable problem (by due attention to controlling prior inputs,
and/or to invariant correlations between prior and present events) and not the
one of primary concern here. "Intrinsic variability" is intended to refer to
changes in the state of the nervous system which occur largely or entirely for
reasons independent of input signals at both present and prior times.
With this understanding, intrinsic variability can be conceived to take one
of three forms: orderly, deterministic but ill-mannered, and probabilistic.
Circadian rhythms, and other periodic phenomena known to be present in the
nervous, neuroendocrine, and neuromuscular systems, provide clear examples of
the first kind of intrinsic variability. Widespread, coordinated changes in
the state of the nervous system on a roughly twenty-four hour cycle can persist
for long periods of time in the absence of any driving variations in the input
pathways of the nervous system, and there is every reason to think would occur
in the absence of any experience at all with daily variations in input. They
can occur as well in isolated parts of the nervous system, of various sizes
down to the level of individual neurons. Periodic heartbeats are similarly
largely independent of driving input at any point in time, and characterize the
whole heart, as well as parts down to the level of single cells. Though not
directly relevant to the concerns of this article, it is worth noting that, in
both cases, the periodic intrinsic variability at the highest level of
organization is not explained entirely by that at the lowest but depends as
well on the interactions among those elements. Isolated heart cells may beat
at frequencies quite different from each other and from that of the heart as a
whole, which reflects not only the intrinsic properties of the individual heart
cells but the interactions among them as well. Noteworthy too is that both
heartbeat and daily activity cycles can indeed be affected by input signals.
Orderly ntrinsic variability, indeed intrinsic variability in general, is a
contributor to behavior, not an explanation of it.
There is a useful, if somewhat fuzzy, border between "orderly" and
"deterministic but ill-mannered." Periodic intrinsic variability is
well-studied at cellular and molecular levels, and raises no spector of the
supernatural whatsoever. It also provides an equally safe route into the
terrain of the "deterministic but ill-mannered". Any time varying signal,
periodic or aperiodic, can be approximated as closely as one wants by
the sum of a sufficiently large number of periodic signals (sine waves, for
example) of different wave-lengths, amplitudes. and phase relations to one
another. In particular, by summing the signal from a sufficient number of
different elements, one can generate a signal (corresponding, for example, to
the internal state of the nervous system) which is equivalent to the output of
a random number generator. The resulting behavior would be deterministic, in
the sense that it could be predicted if one had a knowledge of the underlying
periodic elements. It can nonetheless be very "ill-mannered" in the sense that
an outside observer, lacking a knowledge of the mathematics necessary to
analyze the observed output, would see the behavior as highly or completely
unpredictable. The category of "deterministic but ill-mannered" includes as
well the various phemonena increasingly and popularly known as "chaotic." It
is now well recognized that the behavior of even relatively simple systems
involving small numbers of interacting non-linear elements (such as neurons)
can be highly unpredictable. Such systems are unpredictable but determinate,
in the sense that the time evolution of their activity can be calculated
algorithmically given known starting conditions. The observed activity pattern
itself, however, may, like that of summed periodic elements, exhibit little or
no discernable order. An additional noteworthy feature of chaotic systems is a
"strong dependence on initial conditions". Very small changes in the starting
conditions can lead to very large differences in subsequent behavior. What
this means is that chaotic systems, while determinate in principal, may not be
so in practice. An observer is inevitably limited to a particular level of
precision in the measurement of starting conditions, and the range within which
the actual starting condition falls may well be early enough to yield a range
of enitrely different outcomes.
In popular parlance, events are either predictable or unpredictable, with
the former corresponding to "orderly" and the latter to "random" or
"probabilistic." The recognition of the category of "deterministic but
ill-mannered" is a generally important intellectual development which, however,
seriously muddies this neat dichotomy. "Orderly", as already described, grades
into "deterministic but ill-mannered," and there is some tendency to equate the
the second with "unpredictable" in general. In fact, there is an important
conceptual distinction between "deterministic but ill-mannered" and random or
probabilistic. The latter terms, at least in their strongest sense, refer to
systems in which it is presumed that the unpredictability is not a matter of
practice but instead one of principal. Quantum mechanics provides the most
explicit modern conceptualization of the intuition that such in principal
unpredictable systems exist; coin flipping or dice-throwing are vernacular
metaphors for the same intution that some phenomena have a truly indeterminate
character. It is not at all clear at the moment whether there exists an
operational criterion by which an observer can distinguish between
variabilities which are "deterministic but ill-mannered" as opposed to
probabilistic. Nonetheless, the distinction is conceptually clean, and has at
least one quite important general implication, as discussed further below.
Orderly intrinsic variability is a well documented aspect of nervous system
function. So too is less orderly intrinsic variability in neuronal activity,
falling into the general area of the second two classes mentioned. There
exists an older literature describing "noise" in both single neuron activity
and in such monitors of global activity as the EEG (such observations were not
always done under conditions adequate to rigorously demonstrate that the
activity was intrinsic, but such is very much consistent with the trend of the
evidence). More recently, evidence of chaotic activity has been obtained in a
variety of systems, including invertebrates as well as the olfactory bulb and
neocortex of mammals. While there are varying degress of uncertainty (both
practical and conceptual) in these observations, there is no question but that
a substantial degree of unpredictability characterizes the physiological
activity of many parts of all nervous systems under conditions of substantially
controlled input activity.
V. The analysis and significance of intrinsic variability
Preceding sections have provided neurobiological evidence in support of the
notion that the Harvard Law of Animal Behavior may reflect actual variability
in nervous system function, rather than supernatural forces or the failure of
investigators to adequately control experimental circumstances. What remains
for discussion is whether this understanding yields productive new lines of
investigation. Two questions are at stake. One is whether there are ways to
usefully explore observed variability, and the other is whether variability has
any functional significance.
V.1. Methodology: a case study in frog and leech
Figure 3A provides a concrete illustration of the Harvard Law of Animal
Behavior, drawn from studies of frog prey orienting behavior. In each of a
series of trials, a live prey item was confined in a cup at a fixed angle and
distance from a given frog. The frog oriented accurately on all trials, in the
sense that a line corresponding to the midsaggital plane of the head passes,
when extended outward, through the cup containing the prey item. The actual
position and head angle, however, varies substantially from trial to trial, as
shown in the figure. Most people's presumption is that such variability
relates to a trial to trial variability in the input signals, an obvious
possibility being a variation in those related to prey location (which can
vary somewhat within the cup) or in the frog's posture at the time the stimulus
was presented. The variability is, however, neither decreased by decreasing
the size of the cup (Figure 3B), nor increased by causing the frog to adopt a
wider than normal range of initial postures (Figure 3C). These findings do not
preclude some more elaborate explanation of the observed variance in terms of
variations in input signals. They do, however, provide the kind of evidence
(equally attainable in humans) which would support further consideration of the
likely existence of intrinsic variability, and motivate additional experiments
(of the kind discussed in previous sections where possible) to try and
rigorously establish its existence.
Even given such a demonstration, many people might discount the
significance of the observed variability, on the grounds that the nervous
system is a complex physical device and that a certain imprecision in the
function of any such device is inevitable. That the observed variation has an
organized character (the frog is always pointing at the prey item) rather than
being random provides some argument against this interpretation but an even
stronger case can be made (here and in other situations). The observed
variability is not constant, but instead varies depending on the circumstances.
In a more complex visual environment (Figure 3D), the variability is less.
This clearly indicates that the variability evident in Figure 3A is not simply
that inherent in the limits of precision of a physical device, but rather an
independently variable property of the physical device.
The same is true for the rigorously verified intrinsic variability
described in an earlier section for the leech. Figure 2C illustrates the
stimulus/response curve for the isolated nervous system, showing that over a
substantial range of stimulus voltages, the response in terms of the presence
or absence of swimming is unpredictable (response percentages neither zero nor
100%). The behavior is much more predictable when the head ganglion is removed
from the nervous system, as show in Figure 2D. For each stimulus voltage the
preparation reliably swims or fails to swim. Here too the intrinsic
variability is not an expression of a constant inevitable imprecision but
rather a characteristic which can be reduced under certain circumstances.
The findings in the leech further suggest that different regions of the
nervous system (different intermediate size boxes) may play different roles in
intrinsic variability, with some areas increasing it and other areas decreasing
it (removal of the tail ganglion in the leech tends to increase observed
variability). Interestingly, the history of neurobiological research, viewed
in this light, suggests a similar conclusion. Experimental neuroscientists,
interested in stable and predictable preparations in which to explore the
details of particular pathways, long ago discovered that removal of the more
rostral parts of the nervous system yields more reliable input/output relations
in the spinal cord. While it is possible that this increased stability results
from elimination of sensory input paths associated with the more rostral parts
of the nervous system, it is equally possible that the observed reduction in
variability would also be observed in a totally isolated nervous system. This
possibility seems worth serious attention, and may relate to the experience of
skilled athletes, whose performance at times seems to depend on blocking out
not only input signals but their own thinking as well.
The general problem of how the nervous system is organized so as to
potentially and somewhat unpredictably associate a given input with any of a
number of outputs similar in overall objective but differing substantially in
detail is a wide open area for new investigation. Related to this are a series
of open questions about the nature of the variability itself (some form of
orderly but ill-mannered, or statistical?) and how it is created (molecular
noise or something at a higher level of organization?). Clearly, an inquiry
proceeding from a presumption of the likely existence of intrinsic variability
is not only both achievable and opens new lines of scientific investigation,
but can usefully bridge between the later and folk psychology as well.
.
V.2. Significance
The physicist Erwin Schrödinger, in his classic What is Life?,
remarked fifty years ago on the remarkable stability and predictability of
biological systems, noteworthy not only in comparison to the thermal noise of
all physical systems but also because the biological systems are, of course,
made up of noisy physical elements. From Schrödinger's perspective the
problem of biological systems was one of how to combine physical elements so
that their inevitable noise was reduced in the combination. The discussion to
this point suggests that aspects of nervous system organization may well have
evolved with exactly the opposite tendency: to enhance some kinds of noisiness
or variability. There is, in fact, some reason to believe this is true of
biological systems in general. If intrinsic variability is not actually the
residual noise of imperfect machinery, the question to be treated at this point
is what functions might it serve?
There is, across the sciences generally, an increasing interest in, and
inquiry into the properties of "complex systems." Associated with this is the
emergence of new and useful metaphors to try and imagine how properties like
those displayed by the brain can emerge from simpler elements. "Complex
systems" in general display substantial form and organization in the absence of
anything corresponding to either an external planner or an internal blueprint;
their properties instead emerge simply from the interactions of large numbers
of elements. In this sense, neither the "machine" explanatory metaphor of the
nineteenth century nor the "computer" metaphor of the later part of the
twentieth is appropriate.
A more useful metaphor for "complex systems", suggested by some workers in
the area, is that of a sandpile, as it might emerge from a continuous flow of
sand dropping onto an elevated dish. The sandpile will grow until it reaches a
particular size and shape. Despite the absence of any planner for the
sandpile, either external or internal, its subsequent size and shape will
persist more or less unchanged, as comparable amounts of sand enter and leave
the sandpile by falling off the edge of the dish. The sandpile is an "open"
and "self-organizing" system. Its form depends on a continuous flow of matter
and energy through it, and emerges entirely from the interaction of large
numbers of elements. Such systems in general display both homeostatic and
adaptive properties. If the pile is disturbed by flattening it, it will, over
time, return to its original form. If the dish is made larger, the sandpile
will, over time, modify its shape and form to make use of the additional space.
Most importantly, in the present context, such open, self-organizing systems
are in general "noisy." While the overall size and shape of the sandpile is
constant, there are continous and substantially unpredictable small changes in
its form: at one point the addition of a few grains of sand results in a
comparable number of grains leaving the pile, at another point the same few
grains added triggers a major avalanche. Despite the appearance of relatively
constant form, the sandpile is in fact constantly in a state of flux, indicated
by its noisiness, and it is in fact precisely because of this flux that the
sandpile displays both homeostatic and adaptive properties.
The sandpile is an instance of a "determinate but ill-mannered system", one
in which substantial unpredictability in practice emerges as a property of the
quite determinate interaction of quite determinate elements. It is a
particularly simple example of such a system, having only one relatively stable
form. More interesting, but not enormously more elaborate, systems can have
many relatively stable forms, and shift among them relatively unpredictably
(imagine an appropriately sized hole in the center of the dish, so that every
once in a while, the pile collapses in the center, changing from a mountain
into a volcano). Nonetheless, the sandpile serves to illustrate an important
general point: that variability is not only not inconsistent with either
homeostasis or adaptability but in fact reflects precisely those phenomena
(large numbers of interacting elements in an open system) which gives complex
systems their adaptive and homeostatic characteristics. Phenomena of both
perception and movement control are currently being usefully explored from this
perspective. Object recognition, for example, may be best thought of as a
variability-added search for a stable configuration of a large number of
interacting elements (two such roughly equivalent stable points corresponding
to the skull and woman with mirror of Figure 1).
There are, however, additional and more direct possible significances of
"noisiness" which should also be born in mind. Game theory illustrates one
such. The success of an organism at any given task is frequently dependent on
the playing out of interactions with other organisms, interactions in which the
behavior of each influences the behavior of the other. If the nature of the
task is such that greater success of one organism necessarily implies lesser
success of the other, then there exists an optimal strategy for the "game", a
well-defined set of rules which will assure the maximal achievable score,
assuming both participants behave "rationally." In many cases (poker providing
one example), it can be proven that this optimal strategy includes deliberate
randomization of behavior. Variability in the frog's prey orienting movements
may, for example, make it more difficult for the prey to successfully escape.
Variability can also be used to achieve an objective in the absence of
existing information about exactly how to achieve that objective. Here the
chemotactic behavior of many protozoans is instructive. In the absence of a
chemical gradient of an attractive substance, these organisms move continuously
with intermittent and apparently random changes of direction. The frequency of
direction changes is, however, a function of the concentration of attractive
substances, decreasing with higher concentrations. Because of this, an animal
in a concentration gradient will reliably end up at the source, always, of
course, by a different path. There is some evidence that frogs, once they
acquire information about target location, may have a related capability to use
their variability to create novel ways of reaching that target when more
straightforward methods are unavailable. Indeed such a "creative" use of
available variability may be a quite general property of movement control
systems which display "motor equivalence", and perhaps contribute to explaining
how it comes to be that years on a playground basketball court can yield
performances never before seen.
The discovery, exploration, and creation of novelty has, of course,
significance beyond the restricted domain of movement control, and in this
wider terrain too variability plays obvious roles. "Noise" is an important
element in artificial intelligence systems, where it is used to prevent
premature settling on a less good solution to a task so as to ensure the search
continues until better ones are found. Most humans have similar experience
with needing to "get out of a rut" in their own day to day lives. More
generally, substantially unpredictable variability must necessarily underlie
all genuinely creative processes, since by definition they represent ways of
dealing with phenomena for which the underlying rules are unknown. Dreams
provide a familiar archetype of this kind of variability: while the elements in
dreams (blue, three horns, snake, woman's head) are frequently familiar, the
particular ways they are put together (a blue three-horned woman's head on a
snake's body) frequently display a substantially unpredictable variability.
The experiences one has with such things represent useful (and used)
explorations of the novel, and the unpredictable combinations themselves
sometimes emerge as quite significant insights (the "Eureka!" phenomenon).. .
The preceding is intended to establish that variability is by no means the
residium left over when everything else is understood, but may instead in many
cases be part of the very essence of what is to be explained, since many
aspects of behavior relate more or less directly to problem solving,
exploration, and creation. It is not intended to be a complete catalogue of
the possible significances of intrinsic variability, and the issue of whether
one is dealing with "determinate but ill-mannered" as opposed to probabilistic
has deliberately been left vague for the cases discussed. If there is in fact
a real distinction between a sufficiently chaotic process and a random number
generator, it is probably not a distinction which matters for the situations
described (though this is in fact a matter worth exploration). There is,
however, one additional phenomenon worth mentioning in a discussion of
variability where the distinction is in fact critical.
The existence and nature of "free will" is a major unsettled issue in
western philosophy, one for which the Havard Law of Animal Behavior is
relevant. Roughly speaking, the question is whether the output of a given
organism is under its own control, as opposed to being fully determined by
in-principle-knowable causal factors. In neurobiological terms, one component
of the question can usefully be rephrased in terms of whether complete
information about the genome of an organism, together with similarly complete
information about all input signals which the organism has received during its
lifetime, would suffice to make the next output predictable. The discussion of
intrinsic variability given to this point implies that the answer to this
question is no: even identical twins reared under identical circumstances would
behave differently if a sufficiently chaotic intrinsic variability influenced
their output, either directly or via the longer lasting effects of intrinsic
experiences associated with dreaming or thinking. And they would certainly
behave differently if the relevant intrinsic variability was truly
probabilistic. This frees one to entertain the possibility that behavior is
not fully determined. Finding oneself behaving unpredictably does not,
however, fully satisfy the notion of "free will." The remaining question is
that of control. Does an individual organism in fact have any control over its
output?
A reconsideration of Figure 1 suggests that some organisms, at least, can
in fact exert meaningful control over their own output. As earlier described,
the figure can, somewhat unpredictably, be seen as arrows pointing either to
the right or to the left . Most individuals, staring at the figure for a
sufficiently long time, can in fact see it at one time as one and at another
time as the other. It is also possible for humans (at least) to withhold
output (in response, for example, to the question "what do you see?") until the
picture is seen as one or the other of the two images. One is, in short,
capable of monitoring the consequences of the intrinsic variability of one's
own nervous system, and withholding an output until that variability yields
something with which one is satisfied. Individual organisms, including
individual humans, are both made up of and are a part of complex causal webs of
interacting elements. That they are fully under their own control is clearly
not the case. The present discussion, however, suggests that the combination of
intrinsic variability and a monitoring/editing function may indeed endow at
least those organisms which possess the combination with a quite satisfying
degree of "free will."
VI. Conclusions
The Harvard Law of Animal Behavior clearly warrants more serious attention
than it usually receives. A reasonable skepticism about observed behavioral
variability, due to the possible existence of unknown but identifiable
uncontrolled variables, is both reasonable and demonstrably productive as a
research strategy, insofar as it results in the identification of new causal
influences. At the same time, the phenomena of intrinsic variability discussed
here provide viable alternative explanations of behavioral variability,
potential explanations which are at least as likely, rigorously explorable, and
significant as an unknown uncontrolled but potentially controllable variable.
Two more general implications of this conclusion are worth making
explicit. The first is that the equation between "understanding behavior" and
"being able to predict the behavior that a given organism will display at a
given time", no matter how obvious it might seem nor how useful in particular
inquiries, is fundamentally flawed as a general prospectus for the behavioral
and biobehavioral sciences. Some degree of unpredictability is, as suspected
by the common sense view, a fundamental aspect of behavior. This raises some
interesting questions about the meaning of "understanding" in the context of
the scientific study of behavior, but other disciplines, notably physics, have
managed to incorporate "uncertainty" into their prospectuses, and there is no
reason to believe that the scientific study of behavior can not successfully
incorporate a "biobehavioral uncertainty principle" as well. Acknowledgement
of such a principal would, in any case, contribute to a rapprochment between
the common sense and scientific explorations of behavior, reducing the
perceived threat to human dignity associated with a presumption that behavior
(including that of humans) can in principlebe fully predicted, while at the
same time making clearer the successes inherent in being able to say why an
organism behaves the way it does when it does. Intrinsic variability
not only removes the spectre of absoute predictability, but may, as discussed,
provide a basis for admitting more fully into scientific discourse the concept
of free will.
The second, important general implication is that intrinsic variability is
not necessarily "noise", in the sense of being an inevitable byproduct of any
real system, but instead is, in many cases at least, an essential ingredient of
successful behavior. Scientists tend to focus on the "adaptiveness" of
behavior, on the underlying organization which yields an "optimal" behavior in
a given context. What game theory establishes is that, in many contexts at
least, "optimal" is not a particular behavior, but rather variable behavior.
Even more importantly, however, what is forgotten in the emphasis on
"adaptiveness" and "optimality" is that much of behavior is fundamentally
exploratory. Even when the task is known, as in the case of frog orienting,
the underlying neuronal organization seems to be organized so as to explore
various ways of achieving that task. To put the matter differently, "play"
(and its associated behavioral variability) is not purely entertainment or a
luxury to be given up when things get serious. It is itself a highly adaptive
mechanism for dealing with the reality that the context for behavior is always
largely unknown. Given that the adaptiveness of behavior itself derives from
an evolutionary process in which variability and play are absolutely essential
(and that no more successful mechanism for creating adapted systems is known),
this should not be surprising. Scientists have, however, tended to focus more
on the selection process than on the underlying variability without which it
cannot act. In any case, recognition of intrinsic variability in the nervous
system provides a second basis for rapprochment between the scientific and
common sense perspectives on behavior: playfulness is indeed not only to be
enjoyed but to be accorded high value for its fundamental role in the success
of all organisms, including humans.
BIBLIOGRAPHY
Abbs, J.H. and Cole, K.J. (1987) Neural mechanisms of motor equivalence and
goal achievement. In: "Higher Brain Function" (S.P. Wise, ed.), Wiley, New
York, pp 15-44.
Bak, P. and Chen, K. (1991) Self-organized criticality. Scientific America,
January, pp 46-53.
Barlow, G. W. (1977) Modal action patterns. In: "How animals communicate"
(T.A. Sebeok, ed.) Indiana University Press, Bloomington, Indianna, pp
98-133.
Berkinblitt, M.B., Feldman, A.G., and Fukson, O.I. (1986) Adaptability of
innate motor patterns and motor control mechanisms. Behav. Brain Sci. 9:
535-638.
Grobstein, P. (1988) From the head to the heart: some thoughts on similarities
between brain function and morphogenesis, and on their significance for
research methodology and biological theory. Experientia 44: 960-971.
Grobstein, P. (1992) Directed movement in the frog: motor choice, spatial
representation, free will? In "Neurobiology of Motor Programme Selection" (J.
Kien, C.R. McCrohan, and W. Winlow, eds.), Pergamon Press, New York, pp
250-279.
Kauffman, S.L. (1993) "The Origins of Order", Oxford University Press, New
York.
Linsker, R. (1990) Perceptual neural organization: some approaches based on
network models and information theory. Ann. Rev. Neurosci. 13: 257-281.
Llinás, R. (1990) Intrinsic electrical properties of nerve cells and
their role in network oscillation. Cold Spring Harbor Symp. Quant. Biol. 55:
933-938.
Pittendrigh, C.S. (1993) Temporal organization: reflections of a Darwinian
clock watcher. Ann. Rev. Physiol. 15: 17-54.
Poundstone, W. (1993) "Prisoner's Dilemna" Doubleday, Anchor Books, New
York
Ruelle, D. (1991) "Chance and Chaos" Princeton University Press, Princeton,
New Jersey.
Shepard. R.N. "Mind Sights" (1990) W.H. Freeman and Co., New York.
Skarda, C.A. and Freeman, W.J. (1987) How brains make chaos in order to make
sense of the world. Behav. Brain Sci. 10: 161-195.
FIGURE LEGENDS
Figure 1. An ambiguous figure. Created by the author based on a figure by
Roger Shepard in Mind Sights (W.H. Freeman and Company, 1990). The
figure, sometimes seen as a series of arrows pointing to the left and other
times as a series of arrows pointing to the right, provides a striking example
of the importance of an appreciation of intrinsic variability for the
understanding of behavior (see text).
Figure 2. Intrinsic variability in the leech nervous system. A. A record of
continuous electrical activity from one peripheral nerve of an isolated leech
nervous system. The observed activity varies with time despite the absence of
variations in input, and includes periods of activity known to be associated
with well defined output behaviors such as swimming (the bursting activity
beginning at the arrow). B. Variability in response to a fixed stimulus.
Shown are five trials on each of which the same electrical stimulus (line below
bottom trace) was applied to a nerve other than the one being recorded from.
Notice that the same stimulus sometimes, but not always (B4) causes swimming.
Even on trials when swimming does result, there is substantial variability in
the pattern and duration of activity prior to the initiation of swimming , as
well as in the swimming activity itself (B1,2,3,5). C. A stimulus response
curve for the isolated leech nervous system. For each of a series of stimulus
voltages (abscissa) the figure shows the percentage of trials (ordinate) using
that voltage on which a swimming episode resulted. Notice that, over a wide
voltage range, the same stimulus sometimes does and sometimes does not trigger
a swim. D. A stimulus response curve for the isolated leech nervous system in
the absence of the most head ganglia. Conventions as in C. Notice that there
is now a sharp transition between stimulus voltages which reliably do and do
not trigger swimming episodes. Unpublished studies by Peter Brodfuehrer and
Paul Grobstein.
Figure 3. Behavioral variability in the frog. A. Base line behavior. On
each of ten trials, the frog was positioned at the lower cross (long vertical
line is body axis, short horizontal line the line between the frog's eyes) and
presented with a live mealworm confined in a cup (short upper horizontal
line.Thinner oblique lines show the position and angle of the frog's head after
the first movement on each trial. Notice that both position and angle vary
substantially. B. Reduction of stimulus cup size (shorter upper horizontal
line). Notice that there is no corresponding reduction in the variability of
final head position. C. Increase in variability of initial posture
(broadening of line corresponding to body axis). Notice that there is no
corresponding increase in the variability of final head position. D. A richer
visual environment. Notice that, under this condition, the observed
variability in final head position significantly decreases. Unpublished
studies by Paul Grobstein, Jeff Oristaglio, and others.
|