Such knowledge has provided a valuable starting point for more detailed
study of olfaction. But it leaves two important issues unresolved. The first
is the classic problem of separating foreground from background: How does
the brain distinguish one scent from all others that accompany it?
"PHASE PORTRAITS" made from electroencephalograms (EEGs)
generated by a computer model of the brain reflect the overall activity of
the olfactory system at rest (above) and during perception of a familiar
scent (right). Resemblance of the portraits to irregularly shaped, but still
structured, coils of wire reveals that brain activity in both conditions is
chaotic: complex but having some underlying order. The more circular
shape of the right-hand image, together with its greater segregation of
color, indicates that olfactory EEGs are more ordered-more nearly
periodic-during perception than during rest.
Also, how does the brain achieve what is called generalization-over-
equivalent receptors? Because of turbulence in nasal air flow, only a few of
the many receptors that are sensitive to an odorant are excited during a
sniff, and the selection varies unpredictably from one sniff to the next.
How does the brain recognize that signals from different collections of
receptors all refer to the same stimulus? Our investigations begin to
suggest answers to both problems.
Many of our insights were derived from intensive studies of the olfactory
bulb. Those experiments show clearly that every neuron in the bulb
participates in generating each olfactory perception. In other words, the
salient information about the stimulus is carried in some distinctive
pattern of bulbwide activity, not in a small subset of feature-detecting
neurons that are excited only by, say, foxlike scents.
Moreover, although this collective neural activity reflects the odorant, the
activity itself is not determined solely by the stimulus. Bulbar functioning
is self-organized, very much controlled by internal factors, including the
sensitivity of the neurons to input.
The experiments uncovering the collective activity were conceptually
simple. By applying standard reinforcement techniques, we trained
animals, often rabbits, to recognize several different odorants and to
behave in particular ways when they did-for instance, to lick or chew in
expectation of food or water. Before training was started, we attached 60
to 64 electrodes 0.5 millimeter apart in a gridlike array to a large part of
the bulbar surface.
During training and thereafter, the array enabled us to collect sets of 60 to
64 simultaneously recorded electroencephalogram (EEG) tracings as the
animals breathed in and out, sometimes sniffing familiar scents and
sometimes not. Each tracing reflects the mean excitatory state of local pools
of neurons lying in a well-defined layer immediately beneath the
electrodes. Rises in the wavelike tracings indicate increasing excitement;
dips represent diminished excitement caused by inhibition.
The EEGs should not be confused with recordings of impulses fired by
individual axons or by pools of neurons, although each EEG is related to the
firing pattern of neurons in a neighborhood of the cerebral cortex. The
tracings detect essentially the same information that neurons assess when
they "decide" whether or not to fire impulses, but an EEG records that
information for thousands of cells at once.
To better understand exactly what the EEG shows, it helps to know some of
the details of how cortical neurons operate. Such cells continuously receive
pulses-usually at projections known as dendrites-from thousands of other
neurons. The pulses are conveyed at specialized junctions called synapses.
Certain incoming pulses generate excitatory waves of electric current in
the recipients; others generate inhibitory waves [see top illustration on
page 82]. These currents-"dendritic currents"-are fed through the cell
body (which contains the nucleus) to a region called the trigger zone, at the
start of the axon.
There the currents cross the cell membrane into the extracellular space. As
they do, the cell calculates the overall strength of the currents (reflected in
changes in voltage across the membrane), essentially by adding excitatory
currents and subtracting inhibitory ones. If the sum is above a threshold
level of excitation, the neuron fires.
The mechanism producing each EEG tracing similarly sums the currents
initiated at the dendrites, but it taps the currents after they leave the cell.
The tracings reflect the excitatory state of groups of neurons rather than of
individual ones, because the extracellular space is traversed by currents
from thousands of cells.
BASIC FLOW OF OLFACTORY INFORMATION IN THE BRAIN
In our experiments the EEG tracings from the
electrodes in an array are as unpredictable and irregular as freehand
scrawls. Yet they manifest perceptual information.
In living individuals, EEGs always oscillate, or rise and fall, to some extent,
but the oscillations are usually quite irregular. When an animal inhales a
familiar scent, what we call a burst can be seen in each EEG tracing. All the
waves from the array of electrodes suddenly become more regular, or
ordered, for a few cycles-until the animal exhales. The waves often have a
higher amplitude (height) and frequency than they do at other times.
The burst waves are often called 40 hertz waves, meaning that they
oscillate at about 40 cycles per second. Because the frequency can actually
range from 20 to 90 hertz, I prefer to call them gamma waves, in analogy
with a range of high-frequency X rays.
The fact that the bursts represent cooperative, interactive activity is not
immediately clear in the EEG plots, because the burst segments differ in
shape from tracing to tracing in a simultaneously recorded set.
Nevertheless, by taxing our computers, we find we are able to tease out
evidence of collective behavior from the complex background.
In each set of burst recordings, we can identify a common waveform, or
carrier wave: a shared pattern of rises and falls that is embedded in each
tracing. The average amplitude is not identical across the set-some
versions of the carrier wave are shallow, and others are deep. But all of
them curve up and down nearly in synchrony. The common behavior
makes up between one quarter and three quarters of the total activity of
the neurons giving rise to each trace.
Curiously, it is not the shape of the carrier wave that reveals the identity
of an odor. Indeed, the wave changes every time an animal inhales, even
when the same odorant is repeatedly sniffed. The identity of an odorant is
reliably discernible only in the bulbwide spatial pattern of the carrier-
wave amplitude [see top illustration on page 84].
Amplitude patterns become especially clear when we plot the average
amplitude of the individual versions of the carrier wave on a grid
representing the surface of the bulb. The resulting "maps" resemble
contour diagrams that indicate the elevations of mountains and valleys. As
long as we do not alter the animals' training, the same map emerges every
time an animal sniffs a particular odorant, even though the carrier wave
differs with each sniff.
These maps have helped demonstrate not only that perception requires
global bulbwide activity but also that the bulb participates in assigning
meaning to stimuli. The amplitude map representing a given odorant
changes strikingly when we alter the reinforcement associated with that
scent. If the bulb did not bring experience to bear on perception, the map
would remain constant even after the conditioned association had been
changed.
NEURONS OF THE OLFACTORY SYSTEM share information through a rich web of synapses, junctions where signals flow from neuron to neuron.
Usually signals pass from projections called axons to projections called
dendrites, but sometimes they pass from dendrite to dendrite or axon to
axon. The widespread sharing leads to collective activity. In this highly
schematic diagram, red shading signifies that a neuron is exciting another
cell, black shading that a neuron is inhibiting another.
We believe that something we call the nerve cell
assembly is both a crucial repository of past associations and an essential
participant in the formation of the collective bulbar burst. The hypothetical
assembly consists of neurons that have simultaneously been excited by
other neurons during learning.
More than 20 years ago my colleagues and I discovered that when animals
are trained by reinforcement techniques to discriminate olfactory stimuli,
certain synapses that connect neurons within the bulb and within the
olfactory cortex become selectively strengthened during the training. That
is, the sensitivity of the postsynaptic cells to excitatory input-a property
known as gain-is increased at the synapse, so that an input generates a
greater dendritic current than it would have generated in the absence of
special training. Technically, gain is the ratio of output to input-here, the
net strength of the dendritic currents to the number of incoming pulses.
The strengthening occurs not in the synapse between an input axon (such
as a receptor from the nose) and the neuron it excites (such as a bulbar
neuron) but in the synapse between connected neurons that are
simultaneously excited by input neurons during learning. Neurons in the
bulb and in the olfactory cortex are connected to many others in those
regions.
EEG WAVES reflect the mean excitation of pools of neurons. Excitatory inputs at synapses generate electric currents that flow in closed loops
within the recipient neuron toward its axon, across the cell membrane into
the extracellular space and, in that space, back to the synapse (red arrows).
Inhibitory inputs generate loops moving in the opposite direction ( black
arrows). In cells the trigger zone adds current strengths (reflected in
changes in voltage across the membrane), and it fires impulses if the sum
is sufficiently positive. Electrodes on the brain tap those same currents
after they leave the cell. The resulting EEGs indicate the excitation of whole
groups of cells, not individuals, because the extracellular avenues from
which the EEGs arise carry currents contributed by thousands of cells.
Such strengthening is predicted by the widely accepted Hebb rule, which
holds that synapses between neurons that fire together become stronger,
as long as the synchronous firing is accompanied by a reward. (The
strengthening is now known to involve "modulator" chemicals that the
brain stem releases into the bulb and cortex during reinforcement.)
We infer from our data that a nerve cell assembly, consisting of neurons
joined by Hebbian synapses, forms for a particular scent as an individual is
reinforced for learning to identify that odorant. Thereafter, when any
subset of neurons in the assembly receives the familiar input, the entire
assembly can rapidly become stimulated, as excitatory signals speed across
the favored Hebbian synapses. The assembly, in turn, directs the rest of
the bulb into a distinct pattern of activity.
If we are correct, the existence of a nerve cell assembly would help explain
both the foreground-background problem and generalization-over-
equivalent receptors. In the first instance, the assembly would confer
"frontrunner" status on stimuli that experience, stored in the Hebbian
synapses has made important to the individual. In the second instance, the
assembly would ensure that information from any subset of receptors,
regardless of where in the nose they were located, would spread
immediately over the entire assembly and from there to the rest of the
bulb.
SIMULTANEOUS RECORDINGS from the olfactory bulb (a) and front (b) and rear (c) parts of a cat's olfactory cortex show low-frequency waves
interrupted by "bursts"-high-amplitude, high-frequency oscillations that
are generated when odors are perceived. The average amplitude of a burst
is some 100 microvolts. Each lasts a fraction of a second, for the interval
between inhalation and exhalation.
As important as the nerve cell assembly is to
perception, it does not by itself generate bulbwide bursts of collective
activity. For a burst to occur in response to some odorant, the neurons of
the assembly and the bulb as a whole must first be "primed" to respond
strongly to input.
Two important processes complement the priming accomplished by the
development of Hebbian synapses. Both processes affect the gain, doing so
by altering the sensitivity of the trigger zones, not the synapses. Here the
gain is the ratio of the number of pulses fired (output) to the net dendritic
current (input). The total gain is the product of the gain at the synapses
and trigger zones.
One primer is general arousal. Our experiments show that the gain in
neuronal collectives increases in the bulb and olfactory cortex when an
animal is hungry, thirsty, sexually aroused or threatened [see illustration
on page 85]. Such priming seems to be accomplished by axons from
elsewhere in the brain that release modulatory chemicals (other than those
involved in forming Hebbian synapses).
The other primer is input itself. When cortical neurons are excited, their
output increases. Each new input they receive while they are still excited
raises their output markedly, indicating that their gain has been increased
by the input. This increase occurs over a particular range of input. If the
net input is strongly inhibitory, no pulses are fired. Above some very high
level of excitatory input, neurons fire at their maximal rate and cannot do
more, even if the input is increased. In the wide range between, however,
pulse output increases along a sigmoid (S-shaped) curve. The steepness, or
slope, of the curve reflects the gain.
The discovery of an increase in gain with excitation is particularly
noteworthy because most neural network models assume neurons are at
maximum gain when they are at rest. Both excitation and inhibition are
generally assumed to decrease gain, so that the networks constantly
maintain stability. Such assumptions are inappropriate for the brain
because they do not allow networks to generate explosive changes.
Hence, it seems that information from odorants is fed by a small number of
receptors to a still smaller number of cells in the bulb. If the odorant is
familiar and the bulb has been primed by arousal, the information spreads
like a flash fire through the nerve cell assembly. First, excitatory input to
one part of the assembly during a sniff excites the other parts, via the
Hebbian synapses. Then those parts reexcite the first, increasing the gain,
and so forth, so that the input rapidly ignites an explosion of collective
activity throughout the assembly. The activity of the assembly, in turn,
spreads to the entire bulb, igniting a full-blown burst.
The bulb then sends a "consensus statement" simultaneously along parallel
axons to the olfactory cortex. What must next be made clear is how that
cortical area distinguishes the consensus statement from the background of
other stimuli impinging on it from the bulb and elsewhere.
WHY EEG WAVES OSCILLATE
Alternating rises and falls in amplitude stem from negative-feedback
circuits that are established by the interaction of pools of excitatory and
inhibitory neurons. When the pools have been sensitized to input, even a
small input can trigger. A burst of high-amplitude oscillation. The diagrams
represent neuronal activity at the end of each quarter cycle. Dark shading
signifies great excitement; lighter shading signifies less excitement.
The answer undoubtedly has to do with the wiring
joining the bulb to the cortex. The bulb generates trains of impulses that
run simultaneously along the parallel axons leading from the bulb to the
cortex. Each axon branches extensively and transmits pulses to many
thousands of neurons across the olfactory cortex, and each cortical target
cell receives input from thousands of bulbar cells.
The carrier activity of the incoming lines, which is synchronized by
cooperation, probably stands out for the simple reason that such signals
add together- nonsynchronous inputs, which are not at the carrier
frequency and phase, effectively cancel one another. Thus, every recipient
neuron in the olfactory cortex picks up a share of the cooperative bulbar
signal and transmits the summed signals to thousands of its neighbors
simultaneously.
In response, the massively connected neurons of the cortex, which have
formed their own nerve cell assemblies, promptly generate their own
collective burst, albeit one having a carrier wave and a spatial amplitude
pattern that differ from those in the bulb. In essence, the transmission
pathway for the global pattern in the bulb launders the bulbar message; it
removes "noise," so that only the collective signal affects the olfactory
cortex significantly. Just as a burst in the bulb guarantees the delivery of a
coherent message to the cortex, so presumably does the global burst in the
cortex enable outgoing messages from that region to stand above the din
when they reach other regions of the brain.
There are many reasons why we believe the activity of the brain both
during and between bursts is chaotic, not merely random. But before I
delve into those reasons, let me clarify further what is meant by chaos.
At the risk of oversimplification, I sometimes like to suggest the difference
between chaos and randomness by comparing the behavior of commuters
dashing through a train station at rush hour with the behavior of a large,
terrified crowd. The activity of the commuters resembles chaos in that
although an observer unfamiliar with train stations might think people
were running every which way without reason, order does underlie the
surface complexity: everyone is hurrying to catch a specific train. The
traffic flow could rapidly be changed simply by announcing a track change.
In contrast, mass hysteria is random. No simple announcement would
make a large mob become cooperative.
One of the most convincing early clues to the presence of chaos was an
aperiodic (nonrepeating) common carrier wave everywhere in the bulb not
only during bursts but also between bursts-even when there was no
extrabulbar stimulus driving that collective activity. The lack of external
driving meant the activity was self-generated by the bulb. Such self-
organization is a characteristic of chaotic systems [see "Chaos," by James B.
Crutchfield, J. Doyne Farmer, Norman H. Packard and Robert S. Shawl
SCENTIFIC AMERICAN, December 1986].
Another clue was the apparent ability of neural collectives in the bulb and
cortex to jump globally and almost instantly from a nonburst to a burst
state and then back again. Rapid state changes are called phase transitions
by physicists and bifurcations by mathematicians. Whatever they are
called, dramatic changes in response to weak input are, it will be recalled,
another feature of chaotic systems. Bifurcation is significantly harder to
control in random systems.
We gained more evidence for chaos by developing computer models of the
olfactory system as a whole: the bulb, the cortex, the connections between
them and the input to both areas from outside the system. We simulated
the activity of the system by solving sets of ordinary differential equations
that describe the dynamics of local pools of neurons.
First we demonstrated that the model did in fact represent the olfactory
system accurately. With no more than a single pulse (equivalent to
excitation of a few receptors) to start the system, the model sustained
activity that closely resembled aperiodic olfactory EEGs.
After we " trained" the model to recognize specific odorants, the bulbar
segment generated bursts in response to the selected inputs, and the
embedded common carrier waves yielded distinct and consistent
amplitude maps. Moreover, whenever we added a new "odorant" to the
perceptual repertoire of our hypothetical subject, an identifying global
amplitude map was created. At the same time, the other maps changed-as
they should, of course, in a true associative memory system. We had
earlier found such changes in test animals after they were trained to
recognize stimuli beyond the ones they had learned initially.
COMMON CARRIER WAVE emerged from 60 EEGs recorded information
is contained in the spatial pattern of amplitude simultaneously from the
olfactory cortex of a rabbit as it recognized a scent (left). The wave is
nearly the same in each recording, except that the amplitude varies. The
shape of the carrier wave does not indicate the identity of the scent. That
information is contained in the spatial pattern of amplitude across the
cortex, which can be displayed as a contour plot(right), much like the plots
of elevations in topographic maps. The colored contours represent the
highest amplitude; successive contours represent the lower amplitudes.
Our model yielded additional evidence for chaos
when we coaxed it to produce mock EEGs of extended bursts and of
"interburst" activity in the intervals between bursts. Because the artificial
EEGs persisted longer than EEGs normally do, we were able to plot what are
called phase portraits of the predicted behavior of the olfactory system,
both during and between bursts. The portraits can show at a glance
whether the dynamics may be chaotic.
The details of how such portraits are made and why they are a valid
reflection of global activity in the olfactory system are too complex to
discuss at length. Nevertheless, for those readers familiar with phase
portraits, I should note that we plotted the portraits in a three-
dimensional grid and added color to display a fourth dimension. Each axis
represented EEG amplitude from some part of the olfactory system, such as
the bulb or a subsection of the olfactory cortex. A range of colors from red
to blue represented high to low amplitude from a fourth part of the
system.
We plotted as a point one set of three amplitudes, measured at a given
moment. Next we plotted another point from the set, representing a
thousandth of a second later, and connected the two points with a colored
line. Then we plotted a third point, and so on. We rotated the final image in
space to find the most informative point of view.
CONTOUR PLOT at the left emerged consistently from bulbar EEG's of a rabbit that had been conditioned to associated the scent of sawdust with a particular reinforcement. After the animal learned to recognize the smell of banana (middle), however, reexposure to sawdust led to the emergence of a new sawdust plot (right). The change shows that bulbar activity is dominated more by experience than by stimuli; otherwise, sawdust would always give rise to the same plot.
The pictures supported the possibility of chaos, because the images
resembled loose coils of wire having different shapes and color
distributions. If the model olfactory system had behaved randomly, there
would be no coherent shapes, just dots spread everywhere, like "snow" on
a television set. If, on the other hand, the system was predictable in detail,
the shapes would be simpler; they might resemble a spiral, a folded circle
or a torus (a doughnut).
The shapes we found represent chaotic attractors. Each attractor is the
behavior the system settles into when it is held under the influence of a
particular input, such as a familiar odorant. The images suggest that an act
of perception consists of an explosive leap of the dynamic system from the
"basin" of one chaotic attractor to another; the basin of an attractor is the
set of initial conditions from which the system goes into a particular
behavior. The bottom of a bowl would be a basin of attraction for a ball
placed anywhere along the sides of the bowl. In our experiments, the basin
for each attractor would be defined by the receptor neurons that were
activated during training to form the nerve cell assembly.
We think the olfactory bulb and cortex maintain many chaotic attractors,
one for each odorant an animal or human being can discriminate.
Whenever an odorant becomes meaningful in some way, another attractor
is added, and all the others undergo slight modification.
SIGMOID CURVES show the relation between input (wave density) and output (pulse density) at trigger zones in populations of neurons. (The
plots are not valid for individual neurons.) The rising steepness associated
with increased arousal indicates that sensitivity to input-or gain (the ratio
of output to input, or the slope)-rises with arousal. In each case, gain also
increases as neurons that are already excited (those at and to the right of
the circles) receive more stimulation. This input-dependent increase in
gain is essential to the formation of bursts
Identification of chaos does not automatically reveal its source. We suspect
chaos in the brain arises when two or more areas of the brain, such as the
bulb and the olfactory cortex, meet at least two conditions: they excite one
another strongly enough to prevent any single part from settling down,
and, at the same time, they are unable to agree on a common frequency of
oscillation. Competition between the parts would increase the sensitivity
and instability of the system contributing to chaos. Part of the evidence for
the importance of interaction between the bulb and the cortex is that
disconnection of the two regions makes chaos disappear; both parts
become abnormally stable and quiet.
Modulatory chemicals released into the system from elsewhere in the
brain also increase sensitivity to input, both by participating in the
formation of the Hebbian synapses in nerve cell assemblies and by
enhancing arousal. Because various factors maintain great sensitivity, a
very small signal-a whiff, a whisper, a glimpse-can trigger a massive,
collective state change.
Conceivably, the chaos we have observed is simply an inevitable by-
product of the brain's complexity, including its myriad connections. Yet our
evidence suggests that the controlled chaos of the brain is more than an
accidental by-product. Indeed, it may be the chief property that makes the
brain different from an artificial-intelligence machine.
One profound advantage chaos may confer on the brain is that chaotic
systems continually produce novel activity patterns. We propose that
such patterns are crucial to the development of nerve cell assemblies that
differ from established assemblies. More generally, the ability to create
activity patterns I may underlie the brain's ability to generate insight and
the "trials" of trial and-error problem solving.
We have found widespread, apparently chaotic behavior in other parts of
the brain. That finding does not necessarily imply that other sensory
systems operate as the olfactory system does. But we think they do.
Indeed, we and other investigators have documented gamma bursts across
large cortical regions involved in recognizing visual images. As in the
olfactory system, familiar visual stimuli are associated with specific
amplitude maps of common carrier waves. I predict that when people
examine drawings in which foreground and background are ambiguous, so
that perception alternates between two images, the amplitude maps will
be found to alternate as well.
I begin to envision the general dynamics of perception. The brain seeks
information, mainly by directing an individual to look, listen and sniff. The
search results from self-organizing activity in the limbic system (a part of
the brain that includes the entorhinal cortex and is thought to be involved
in emotion and memory), which funnels a search command to the motor
systems. As the motor command is transmitted, the limbic system issues
what is called a reafference message, alerting all the sensory systems to
prepare to respond to new information.
And respond they do, with every neuron in a given region participating in
a collective activity-a burst. Synchronous activity in each system is then
transmitted back to the limbic system, where it combines with similarly
generated output from other sensory systems to form a gestalt. Then,
within a fraction of a second, another search for information is demanded,
and the sensory systems are prepared again by reafference.
Consciousness may well be the subjective experience of this recursive
process of motor command, reafference and perception. If so, it enables the
brain to plan and prepare for each subsequent action on the basis of past
action, sensory input and perceptual synthesis. In short, an act of
perception is not the copying of an incoming stimulus. It is a step in a
trajectory by which brains grow, reorganize themselves and reach into
their environment to change it to their own advantage.
The poet William Blake wrote: "If the doors of perception were cleansed
every thing would appear to man as it is, infinite." Such cleansing would
not be desirable. Without the protection of the doors of perception-that is,
without the self-controlled chaotic activity of the cortex, from which
perceptions spring-people and animals would be overwhelmed by
eternity.
FURTHER READING
MASS ACTION IN THE NERVOUS SYSTEM:
EXAMINATION OF THE NEUROPHYSIOLOGICAL BASIS OF ADAPTIVE
BEHAVIOR THROUGH THE EEG. Walter J. Freeman. Academic Press,
1975.
How BRAINS MAKE CHAOS IN ORDER TO MAKE SENSE OF THE WORLD.
Christine A. Skarda and Walter J. Freeman in behavioral and Brain
Sciences, Vol. 10 No. 2, pages 161-195; June 1987.
THE SYNAPTIC ORGANIZATION OF THE BRAIN. Third edition. Gordon M.
Shepherd. Oxford University Press, 1990.
SYNERGETICS OF COGNITION: PROCEEDINGS OF THE INTERNATIONAL
SYMPOSIUM AT SCHLOSS ELMAU, BAVARIA, JUNE 4-8, 1989. Edited by H.
Haken and M. Stadler. Springer-Verlag, 1990.
Copyright 1991: Walter J. Freeman