Evidence for Information Processing in the Brain

Marc Burock

ABSTRACT

Many cognitive scientists, neuroscientists, and philosophers of science consider it uncontroversial that the brain
processes information. In this work we broadly consider the types of experimental evidence that would support
this claim, and find that although physical features of specific brain areas selectively covary with external
stimuli or abilities, there is no direct evidence supporting an information processing function of any particular
brain area.

scientific process, which includes attention, perception, data

1. INTRODUCTION interpretation, memory, and scientific communication, is
influenced by theory. Perhaps most relevant to our work is
Many cognitive scientists believe that the brain processes Popper’s warning that “…observation statements and statements
information. Bechtel and Richardson (2010), as philosophers of of experimental results, are always interpretations of the facts
cognitive science, consider it uncontroversial that cognitive observed…they are interpretations in the light of theories (italics
scientists involved in neuroimaging research believe that “the in original, p. 90).” Of course, none of this need imply scientific
brain contains some regions that are specialized for processing relativism, and relativism is not assumed in this work.
specific types of information (p241).” Neuroscientists too claim One may presume that only cognitive scientists are qualified
that “the principle function of the central nervous system is to to interpret the experimental evidence in the field. While an
represent and transform information (deCharms and Zador 2000, expert’s assessment carries more weight than the non-involved
p613).” Given such wide-spread acceptance of a belief, it is observer, it is reasonable that anyone who takes time to
appropriate to ask for the justification of this belief. If the understand the evidence and its methods of acquisition is in a
justification is empirical and experimental, then we should look position to construct an interpretation. The force of the
to the research reported by working scientists in the field; if it is interpretation should be based upon the reason of the argument
metaphysical, then we should look to the arguments of and not only its source. Nonetheless, we have performed some
philosophers and theoreticians. of the types of experiments that we are now interpreting.
We will no doubt discover both kinds of justification if we Cognitive scientific evidence, especially neuroimaging
look for it. Yet we assume that cognitive scientists, when stating evidence, has been increasingly subjected to criticisms. To better
that the brain processes information, are primarily stating an demarcate our position, we highlight that we are not specifically
empirical fact or a widely agreed-upon scientific proposition that arguing between distributed versus localized processing in the
is supported by a body of experimental evidence. Like the brain (Utel 2001, Hardcastle and Stewart 2002), and we are not
physicist who can back up the proposition “protons have spin” pointing out the previously discussed technical-methodological
with a presentation of the experimental evidence, we expect that limitations of brain assessing technologies (Logothetis 2008,
the cognitive scientist should be able to do the same regarding a Roskies 2007, Klein 2009). We do share with these authors the
statement about the brain. If the cognitive scientist cannot do broader concern for interpretations of evidence in the field of
this, then the proposition is non-empirical or unscientific. We cognitive science, and how theoretical assumptions influence
are not suggesting a definition of science or solving Popper’s interpretations of evidence, ultimately ending in statements made
demarcation problem, but we are appealing to the belief that by cognitive scientists that carry the weight of scientific fact.
accepted scientific statements are associated with experimental These facts, in turn, are used by naturalistic philosophers of mind
evidence. Without associated evidence, a proposition cannot be to constrain philosophical theory and argument.
scientific.
Our task here, however, is somewhat more involved than an 2. COGNITIVE SCIENCE EVIDENCE TYPES
objective review of the scientific literature. As we have learned
from philosophers of science over the past century, “theory The scientific statement that we will consider is Bechtel and
dominates the experimental work from its initial planning up to Richardson’s proposition “the brain contains some regions that
the finishing touches in the laboratory (Popper 1959, p90).” are specialized for processing specific types of information,”
Hanson (1958) and Kuhn (1962) were among the first to direct although, since we are not specifically arguing against
our attention to the theory-ladenness of scientific observation. localization of brain function, we will consider simultaneously
Brewster (2001) extended this position, arguing that the complete the more general proposition P “the brain processes

1
information.” deCharms and Zador say that it is the function of histological examination, molecular analysis, electroencephalo-
the brain to process (represent and transform) information. There graphy (EEG), magnetoencephalogray (MEG), computed
is no philosophical consensus on how to define a natural tomography (CT), positron emissions tomography (PET), single
biological function, and we will assume that processing photon emitted computed tomography (SPECT), structural
information is a natural function of the brain like pumping blood magnetic resonance imaging (MRI), diffusion tensor imaging
is a natural function of the heart. One might argue that (DTI), magnetic resonance spectroscopy (MRS), and others. By
processing information is a natural process and not a natural ability we mean any observable behavior that can be done by an
function, and we will consider both alternatives, at times blurring animal or human, such as the ability to count out loud, to raise
the distinction in this work. one’s arm, to navigate a maze, to write a sentence, to score above
We wish to consider the experimental evidence that justifies chance on a test, etc.
P. There are tens of thousands of papers in scientific journals In external-stimulus/brain-response (ES/BR) studies, the
that may be used as evidence, thus a systematic evaluation of experimenter systematically manipulates a physical feature of an
every paper independently and subsequent integration of the organism’s external environment, and measures temporally
evidence is not feasible. The task must be simplified, but in a coincident properties of the organism’s brain. The response need
way that addresses the initial question. As a first step, we will not occur precisely simultaneous with the stimulus and is
only consider research that involves measuring or manipulating typically extended in time. Edgar Adrian is generally credited
the physical properties of brain. While a study that does not with pioneering stimulus-response studies of nervous tissue. He
involve brain properties may contribute to our scientific was the first to record the electrical activity of single nerve
understanding of the brain, it can only do so indirectly by fibers, and was subsequently awarded a Nobel Prize in 1932 for
prompting theory formation and characterizing behavioral his work. As an example of his prolific research, he isolated an
phenomena. For example, in 1908 Yerkes and Dodson eel’s eye and optic nerve, attached electrodes to the nerve, and
discovered that performance on a task at first increases with recorded the electrical activity to varying lighting situations
increasing arousal and then decreases once arousal levels become (Adrian and Matthews 1927).
too great. They quantified the intuition that ‘stress’ can enhance Brain responses in ES/BR studies are recorded using a variety
performance. This interesting and useful finding may suggest of techniques based upon electromagnetic brain properties,
neurophysiological correlates of performance and arousal, but it including single-unit intra and extracellular recording, evoked
does not experimentally justify any statement about brain potentials, EEG, MEG, and others. Functional MRI (fMRI) is a
function in a direct sense. popular tool used by cognitive scientists to assess brain
We are primarily interested with research that investigates the properties in response to an ES. Proper interpretation of the
relations between physical brain properties and behaviors, fMRI signal itself requires technical background knowledge
abilities, or physical (sensory) contexts. Cognitive scientists, and (Roskies 2007, Logothetis 2008); briefly, the fMRI signal is
philosophers of science, reference the evidence from this consequence of the magnetic properties of blood components
category of research when making claims about brain function. which covary with local metabolic demands. Other common
Relational evidence, as we will call it, can be broken down into techniques in ES/BR studies include PET and optical imaging.
four major categories: When, within a research protocol, a brain-response is
1) structure/ability studies, recorded while the organism is performing a particular task or
2) external-stimulus/brain-response studies, activity, we call this a task/brain-response (T/BR) study. The
3) task/brain-response studies, and form of the T/BR study is similar to ES/BR studies, except the
4) brain-manipulation/behavioral-response studies. process of completing the task is ‘self-directed’ rather than under
In structure/ability (or S/A) studies, researchers relate the complete control of the experimenter. Often T/BR studies
structure or structural states of the brain to the absence or include ES aspects as well. Memory research provides typical
presence of particular behaviors or abilities. Paul Broca (1861) examples. Poppenk et al. (2010), in studying prospective
popularized this type of research with his lesion-deficit, or lesion memory which is described as the ability to act out postponed
study, when he discovered an individual who could only speak intentions at future times, presented a series of visual scenes (ES)
the syllable ‘tan.’ A post-mortem analysis of the person’s brain to subjects and instructed the subjects to either imagine
revealed damaged brain tissue in the posterior part of the left performing an action associated with the visual scene, or to use
inferior frontal gyrus—a region now known as Broca’s area. the scene as a reminder to perform an action the next time the
Thus Broca related the ability to produce fluent speech to the left same scene was viewed. FMRI was used to measure properties
inferior frontal gyrus. The class of S/A studies includes more of the subjects’ brains during the tasks. Notice that although the
than lesion studies since any physical feature of the brain (e.g. experimenter controls the ES and the task command directly, she
patterns of white matter connectivity) may be associated with the cannot control the process by which the subject completes the
absence or presence of specific abilities. task.
The structure in S/A studies refers to the physical structure of Brain-manipulation/behavioral response (BM/BR) studies
the brain as measured by a variety of measuring techniques, the differ from ES/BR and T/BR studies in that physical properties
most basic being gross anatomical observation of brain tissue. of the brain are directly controlled or manipulated while a
Other measuring techniques include, but are not limited to, behavioral response is observed. Technologies used for BM

2
include lesioning, gene-expression, direct current stimulation, that the structural brain area that is correlated with an ability is in
electrode-based deep brain stimulation, transcranial magnetic fact performing a process proximately responsible for the ability,
stimulation, and light-based optogenetic stimulation, among but this is not the logic of S/A studies. The logic of S/A studies
others. Optogenetic studies are a relatively recent advance. is as follows: if a subject with structural pattern S cannot do A,
They are based upon the introduction of light-activated channels but a subject without S can do A, then the ability to perform A
into specific populations of neurons, permitting relatively precise must depend upon S in some way. Even if we accept this logic—
control of action potential generation in live organisms (Zhang et which requires that the subjects are similar in every other way
al. 2007). For example, Wyart et al. (2009) expressed light except S—we cannot logically infer that the function of S is to
sensitive genes within so-called Kolmer-Agdur cells of the perform A.
zebrafish, and then non-invasively manipulated the neuronal It should be clear that attributing function based upon S/A
activity of these cells which modulated the swimming behavior studies is not logically justified. Consider this example. There
of the animal. were many times when my computer, the computer I am using
right now, loses a particular ability that I expect it to have. I
3. INTERPRETATION OF THE EVIDENCE recall a time when I was unable to run programs I typically run,
and other programs began running very slowly or would shut
We wish to determine if research studies from the categories down for no apparent reason in mid-session. The problem turned
of relational evidence discussed thus far justify the scientific out to be a dead CPU fan. Should we say that the function of the
claim P that the brain processes information. Again, we will not CPU fan is to run programs quickly and prevent them from
consider every study; rather, we will start with a typical example shutting down? The abilities in question depended upon the
from each category of evidence and attempt to generalize our spinning of the fan, but the fan does not perform the absent
conclusions to the category. abilities—the function of the fan is to cool down the CPU. The
CPU fan participates in a series of causal interactions that run
3.1 S/A programs when ‘everything is working,’ and we could similarly
argue, given the results of a S/A study, that a structural feature of
Beginning with Broca, what can we infer from his S/A study the brain participates in causal interactions that realize a
of subject ‘Tan’ who could only speak the syllable ‘tan’, and particular ability under certain conditions. This does not entail,
more specifically, how do we clarify the relationship between given a S/A study, that the function of the brain structure is to
brain structure and the inability to produce fluent, complex perform the ability in question.
speech (Broca’s aphasia)? It is clear that Broca’s area lesions How do we relate the structure of a brain area to the function
and Broca’s aphasia are related in the sense that they can occur of that area? In S/A studies, that function is always speculatively
contemporaneously within a single individual. It is also clear inferred from an observed ability. We do not study how the
that Broca’s lesion and Broca’s aphasia need not be dynamic processes of the brain area and its relations to the rest of
contemporaneous, for 85% of patients with chronic Broca’s the organism causally make the ability possible—but this is what
aphasia have lesions in Broca’s area, and only 50-60% of we need to understand if we are to assign a function (or process)
patients with lesions in Broca’s area have a persisting Broca’s to the brain area in question. Broca’s area is more selectively
aphasia. Further, surgical excision of Broca’s area—a brain related to the ability to produce grammatically appropriate
manipulation/behavioral response study—has led only to speech than some other parts of the organism. One may argue
transient mutism followed by recovery of the patient to normal that the imperfect but selective correlation between Broca’s area
(Dronkers 2000). This evidence alone is enough to at least and Broca’s aphasia justifies the scientific claim that Broca’s
challenge claims of understanding the function of Broca’s area. area processes linguistic information, but the ability ‘to process
One can of course speculate on the functioning of a brain area linguistic information’ plays no obvious role in the study. To
given an imperfect statistical correlation between that brain area justify the claim that Broca’s area processes linguistic
and an observed ability, but any claims of knowledge of function information given the evidence of a related S/A study—a claim
are excessive: we have knowledge of the correlation and not the made but many scientists in the field—we would have to assume
function. that speaking grammatically appropriate speech involves
Let us assume, falsely, that Broca’s lesions and Broca’s linguistic information processing by the organism as a whole
aphasia were perfectly correlated in the sense that 100% of (because this is the only way that processes and operations enter
patients with Broca’s aphasia have lesions and 100% of patients into S/A studies. We do not observe brain area processes; we
with lesions have aphasia, for it is possible that other S/A studies observe the functioning organism), and then further identify the
exhibit perfect correlation. From this finding, can we conclude ability to process linguistic information with the function of
that perfect correlation between structural brain states and Broca’s area. But neither of these steps is empirically justified.
particular abilities justifies P? This would at first appear to No one directly measures linguistic information processing in the
depend upon the nature of the ability being studied, but it is not study—we simply observe the form of speech—and the
clear that any S/A study could empirically justify that the brain identification of an organism’s ability with the functioning of a
processes information. In S/A studies, one may try to infer brain brain area is a speculative inference. This does not imply that
processes given a correlated ability. There is a tendency to argue Broca’s area plays no role in the ability to produce fluent

3
speech—we simply do not know what that role is given limited while others only appreciably respond to moving bars. In
evidence and theory. describing these cells, Hubel says that “We feel that we have at
We will not address other S/A studies as the form of our least some understanding of a cell if we can say that its duty is to
argument applies to all studies that attempt to relate brain take care of a 1 degree by 1 degree region of retina, 6 degrees to
structures and organism abilities. In summary, the evidence in the left of the fovea and 4 degrees above it, and to fire whenever
S/A studies is composed of correlations between brain structures a light line on a dark background appears, provided it is inclined
and organism-level abilities. To arrive at claims of information at about 45 degrees (Hubel 1962, p168).”
processing in the brain from S/A studies, one must first infer that The evidence from all of these pioneering ES/BR electro-
the brain structure performs a process or function that directly physiological studies cannot be interpreted without the concept
enacts the organism-level ability in question, but this inference is of selective response. Selective response means, loosely, that
speculative. In a second layer of interpretation, one must identify the cell fires action potentials only when the ‘right’ stimulus is
performing the process or function with information processing, present. Put more rigorously, selective response refers to two
but this identification enters as an assumption independent of the characteristics of neuronal cells: (1) the rate or pattern of firing
evidence. action potentials (the spike train) covaries with specific stimulus
properties, and (2) different cells may respond differently to the
3.2 ES/BR same stimulus. Both characteristics are typically implied when
referring to the selectivity of cells in ES/BR studies. If someone
External-stimulus/brain-response studies address questions of discovered a neuron that exhibited (1), but on subsequent
brain function more directly than S/A studies and are probably research discovered that all neurons exhibited (1) in the same
the largest category of relational evidence. Edgar Adrian, in his way, one would not say that the initial neuron was selective for
pioneering ES/BR research, measured the electrical responses of the stimulus, even though it exhibited selectivity for some stimuli
single sensory cells, such as stretch receptors, while they were among others. As well, the fact that different neurons respond
fixed to particular weights. He observed that a cell’s electrical differently to similar stimuli does not imply (1), since neuronal
responses are in the form of stereotyped action potentials, or responses may be random in response to stimuli. Condition (1) is
spikes, and that the rate of producing spikes increases as the a form of within neuron stimulus selectivity, while condition (2)
weight increases. Thus the rate, or frequency, of spikes during a is a form of between neuron stimulus selectivity. For ES/BR
fixed time period is able to predict the magnitude of the stimulus. studies such as Hubel and Wiesel’s, when an ES is chosen and
These early experiments established that single cell responses controlled by the researcher, we assume that the relation
and stimulus magnitudes may reliably covary with each other. between the ES and BR is causal, as this assumption does not
While magnitudes and intensities are important properties of change our interpretation of selectively, even though we use the
stimuli, they are not the only properties of environmental stimuli term ‘covaries’ which has statistical connotations.
that are relevant to an organism. In general, a stimulus may be We are now in a position to evaluate whether Hubel and
characterized by multiple properties. For example, an auditory Wiesel’s ground-breaking ES/BR studies justify the claim that
stimulus may be described by its intensity, frequency spectrum, the brain processes information. In this case we are asking if
temporal envelope, source direction, source distance, and so on. specific neurons, complex cells of the striate cortex, process or
It is possible that a particular cell responds to one of these carry information. The experimental evidence consists of
properties and not the others, or to some combination of recorded responses of complex cells that demonstrate stimulus
properties, which suggests that a cell may be selective for selectivity in the senses of (1) and (2). It seems that selectivity in
specific properties or features of the stimulus. the sense of (2) does not provide any justification that complex
Barlow (1953) was perhaps the first to clearly demonstrate the cells process information; the fact that different cells respond
feature selectivity of sensory cells (Reike et. al 1999). By differently to the same stimulus suggests only that the cells are
recording the electrical activity of retinal ganglion cells in the different in some way.
frog, he was able to show that the cell’s activity covaries with the Claims of information processing, if they are justified by this
location and size of a circular spot light on the retina. After experiment, must follow from the evidence that complex-cell
systematically varying the light spot’s size and location, Barlow spike trains covary with the properties of visual stimuli, or in
determined that the cell’s receptive field—the collection of causal language, that different visual stimuli cause different
stimulus properties that maximally activated the cell—is a complex cell spike trains. Considering the latter causal language,
circularly symmetric form called a center-surround field. Spots the fact that different causes reliably produce different effects
of light within a small region of the retina activate the cell, but when mediated by the same cell does not appear to justify the
spots of light away from that region inhibit it. claim that the cell processes information, unless one takes that
Hubel and Wiesel (1962) greatly extended Barlow’s work fact to be a definition of information processing itself. Even so,
and discovered cells of the striate (visual) cortex that have this type of causal relationship appears everywhere one looks. A
surprisingly complicated receptive fields. Two of these cell particular pool ball when hit by other balls with different masses
types are the so-called simple and complex cells, which respond and velocities will undergo different effects. The pool ball may
maximally to appropriately oriented bars or slits of light. Some not appreciably move when stimulated by light or sound at
of the cells are relatively insensitive to the location of the bar, typical intensities. The selectivity of the pool ball to acquire

4
different velocities in response to different causal ‘stimuli’ does that generates a message (ES), the message is transformed by a
not appear fundamentally different than the selectivity of a transmitter—a sensory organ of the organism—into a signal
complex cell, especially if the visual stimulus is taken to be a suitable for biological transmission. The spike train (BR) is
space-time collection of photons. assumed to be this signal and the neuron to be the transmission
On closer analysis, there is a difference between the causality channel. These comparisons are reasonable, but the next stage
in the pool ball example and the relation between the ES and BR of the communication model, however, is problematic.
of complex cells. The pool ball example involves direct Communication requires a receiver that performs the inverse
physical contact and an exchange of energy and momentum, operation of the transmitter, or something that reconstructs the
while the causal response of the complex cell is more indirect. environmental message from the spike train signal.
Photons travel through the lens of the eye and are absorbed by The experimental researcher, the one who discovers selective
photoreceptor cells of the retina. Absorption of photons causes correlations between neuronal spike trains and environmental
the release of the neurotransmitter glutamate at synapses onto so- messages (stimuli), often plays the surrogate role of the receiver
called bipolar cells, causing the electrical field across the or decoder. By describing relational or mathematical mappings
membrane of these cells to become more positive or negative, between the ES and BR, neuroscientists attempt to ‘read the
which respectively increases or decreases the probability of neural code.’ But this is not the sort of information transmission
generating an action potential. Bipolar cells have axons that we were trying to explain. To complete the biological
synapse on other cells, and through a series of neuronal communication model, and to ground information transfer, we
connections, influence the membrane potential of complex cells need to explain how the organism can reconstruct the
and subsequent action potential generation. The causal chain environmental message from its temporal pattern of action
from photons to complex cell response is complicated and likely potentials, and we must demonstrate that the organism
includes causal feedback, yet it is not obvious that a complicated reproduces a similar environmental message within the organism
causal chain is necessarily information processing. itself. The neuronal spike train is not the message—if anything
Even more worrisome is the fact that selective causation need it is the transmission signal or ‘encoded message.’ Although
not imply that the BR has any functional relation to the ES at all. interesting, it is not enough to show that spike trains have the
Nothing rules out the possibility that those selective correlations capacity to represent environmental messages through selective
are accidental—not in the sense that the correlations are covariation. The fact that researchers can mathematically map
statistically spurious, but that those correlations are functionally spike trains back onto stimuli does not say anything about how
irrelevant to the stimuli of interest. As an analogy, suppose my the organism physically reconstructs the environmental message.
computer has a CPU fan with a blue LED light on the fan. The This capacity to map follows immediately from statistical
light, however, is unlit and the fan isn’t spinning. It happens that correlations. Neuroscientists who acknowledge these limitations
when I kick my computer just so on the left side of the front explain that mathematically reconstructing stimuli from spike
cover, the LED lights up, the fan begins spinning but stops after a trains requires taking the homunculus point of view (Reike et al.
second or two, and the light goes out. If I kick it again, just so, it 1999).
starts up for a second then stops. I can reliably cause the fan to For an organism to receive an environmental message, that
turn on for a bit. When I kick the computer in other places, or message must be within the organism and have the same
shake it up, or sing to it, nothing happens to the fan. The fan is structure as the original message. This suggestion may appear
selectively correlated with a specific kick. Perhaps there are radical, but it is simply the completion of Shannon’s
hundreds of computers, constructed at the same factory, that communication model—the same model that supports the
behave similarly. This selective causal relationship does not intuition that the brain processes and transmits information. For
imply that the fan is functionally relevant to my kicking, or example, consider telephonic communication. Air pressure
processes kicking information, or represents kicking. This waves may be converted into analog electronic messages that are
causal relationship may be accidental. Why then, given the encoded into digital signals and transmitted through a physical
evidence of selective responses in ES/BR studies, do many channel. This digital signal, which does not mirror the sound
philosophers and neuroscientists associate information wave in form, reaches a destination where it is reconstructed
processing with this sort of causation? back into an analog message that drives a loudspeaker,
reproducing the original pressure wave. If the original message
3.2.1 Intuition for Information in ES/BR studies was not reproduced (perhaps imperfectly) at a destination, we
could not claim that communication or information transfer took
There is a strong tendency to associate information place. A message is communicated if and only if that message is
processing with the results of ES/BR experiments like Hubel- reproduced at the receiver.
Weisel’s. The spike trains of neurons appear to be relaying If one assumes that the organism receives environmental
specific messages about the external environment to the messages, then in accordance with Shannon’s communication
organism. Claude Shannon (1948), the founder of mathematical model, at least the structure of that message must be physically
communication theory, rigorously defined a model of reproduced within the organism. The alleged encoded
information transfer that may explain this appearance. In message—or spike train—has a physical basis, thus the message
Shannon’s language, the physical environment acts as a source ought to have a physical basis as well. This means that the

5
scientist would have to demonstrate a set of brain-related would agree upon, but that (2) does not obviously follow. The
physical measurements that copy, perhaps imperfectly, the fact that an ES and BR selectively covary, through causal paths,
structure of an environmental stimulus. Let us call this the brain- does not appear sufficient to justify claims of representational
image of an environmental message. It would remain for the content, and it has been argued that covariation of this sort is not
scientist to describe the mechanisms by which neuronal spike even necessary for representational content (Millikan 1989,
trains causally reconstruct the brain-image of a particular Bechtel 1998).
environmental message. We need not expect deCharms and Zador, as neuroscientists,
But no evidence suggests that brain-images exist, so the very to philosophically justify what it means for a spike train to carry
presence of an encoded message within the brain presents a informational content, yet if claims of carrying content do not
problem. In other words, why should the brain contain encoded follow immediately from the observed evidence, then we can
messages that transmit environmental messages, yet never only assume that they are interpreting the evidence or
reproduce the structure of the message itself? The organism communicating the evidence by way of metaphor. But deCharms
requires the actual message, and not only an encoded version of and Zador, along with many other neuroscientists, speak as
it. At this point our analogy to Shannon’s communication model though ‘carrying content’ is a straightforward experimental fact
breaks down. It does not appear that the environment apart from, or in addition to, selective covariation.
communicates a message to the organism, but rather, the From a philosophical perspective, Dretske (1988, 1995)
organism is perhaps translating the environment. Spike trains are argues that regular causal covariation, by itself, implies
not signals corresponding to encoded messages; they are the information carrying. For example, he says that flag poles and
actual messages only in the language of the organism, whatever metal paper clips carry information about temperature because
that might mean. With respect to the organism, the message is the volumes of these metal objects are reliably correlated with
not encoded in anyway, and speaking of a neural code is temperature. But is it not too easy to find this sort of information
metaphorical and at times misleading. The analogy has changed carrying all around us? And why do the objects in question need
from information transfer to language translation. Our goal here, to be regularly or reliably correlated? Any two things that are
however, is not to support other metaphors, but to show that causally related, perhaps probabilistically, transmit the same sort
Shannon’s communication model, which is an integral part of of thing. If the flag pole was hit by a lightning bolt, does not the
modern technology, does not match the relation between an ES flag pole carry information about the energy of the lightning
and BR. bolt?
Neuroscientists, such as deCharms and Zador (2000), So long as the causal relations are understood between objects
repeatedly claim that spike trains carry information or content c and e, then we might say that e carries information about c. If a
about the environment, and suggest what it means ‘to carry situation can be expressed in the form of a law-like equation,
information’: then any parameter on one side of the equation can be said to
carry information about a parameter on the other side equation,
“Imagine recording from the neuron labeled B1 during such as the ideal gas law PV=nRT. If the conditions are
different types of stimuli or behaviors and discovering the probabilistic, then we can use probability theory to derive the
information that this neuron carries about the organism’s distribution of one variable given another, so long as we have
environment—the content of this neuron’s signal (p614- some understanding of the physical connections between
15).” variables. Carrying information, at least according to Dretske,
follows directly from knowing the causal relations between two
In a concrete example about a retinal cell they say that “The physical situations, or from minimally knowing that two
activity of the neuron will be highly correlated with the point of situations are statistically correlated. If by processing
luminance (thus carrying content about this input)(p637).” Like information neuroscientists and philosophers mean that stimulus
in Hubel-Wiesel’s ES/BR experiments, we call this evidence the properties causally or statistically covary with regionally specific
selective covariation between stimulus properties and spike neuronal activity, then we agree with P, although we suggest
trains. deCharms and Zador use the word ‘information’ above to abandoning P in favor of more empirically-grounded statements
possibly mean ‘specific properties or features of the stimulus.’ about covariation.
Given these examples, we can suppose that they would endorse
the following argument: (1) spikes trains and stimulus properties 3.2.2 Other Philosophical Justification of Information in
selectively (and causally) covary, and (2) the (representational) ES/BR studies
content of a spike train is the stimulus property that causes that
spike train. Considering similar ES/BR experimental evidence, Garson
deCharms and Zador do not bring forth any other types of (2003) has attempted to explain a concept of information based
experimental evidence other than selective covariation to justify upon the pioneering electrophysiological ES/BR studies of Edgar
the claim that spike trains carry informational or representational Adrian. Hubel-Wiesel’s and Adrian’s experiments were similar;
content, although they do stress that the representational nature both consisted of presenting stimuli while measuring the
of spikes trains is based upon content and function. We have electrical responses of single cells. Although the technologies,
argued that (1) is a statement about the evidence that all of us organisms, cell types, and stimulus types differed between

6
Hubel-Wiesel and Adrian’s experiments; the evidence in both spike rates map onto differences in intensity, and “not to a
consisted of the relations between stimuli and neuronal spike constant state of the stimulus.”
trains, and it is this evidence that Garson uses to elucidate a We agree with Garson that differences in stimuli are
concept of information. particularly important to the human organism and other animals,
To ground his concept of information, Garson argues—in but Garson does not logically establish differential mapping—
accordance with Adrian—that “differences in the frequency of nor does the scientific evidence primarily support differential
the sequence of action potentials map onto the differences in the mapping—and thus he does reach the goal of deriving a concept
intensity of the stimulus that produce them, and not to a constant of information from the evidence. Even if he did establish
state of the stimulus.” He argues that differences in stimuli and differential mapping, the next step in the derivation, where he
spike trains map to or covary with each other, and that this fact argues that the relation between stimuli and spike trains is
captures the sense in which spike trains and stimuli are arbitrary, meets with difficulties as we will show.
arbitrarily related to each other, thus supporting the Garson provides an alternative way to understand the relation
informational nature of the spike train. between stimuli and spike rates that does not directly depend
Garson’s argument stands or falls with the truth of differential upon the truth of differential mapping. He explains that a given
mapping, yet his belief in differential mapping, which he takes stimulus s1 may be associated with multiple firing rates r1 and r2
from Adrian’s work, does not correspond to the predominant because of adaptation, and that two stimuli s1 and s2 may be
experimental methodology used to acquire evidence in ES/BR associated with a single firing rate r1 for similar reasons. In his
studies. Since Hubel-Wiesel’s experiments, neuronal responses words:
in electrophysiological ES/BR studies are most often understood
by characterizing the feature selectivity of the cell type (Rieke et “For example, suppose a stimulus of intensity si elicits a
al. , p12). This selectively corresponds to the collection of firing rate ri from a neuron. Then si is held constant, and by
stimulus properties that evoke responses for that cell, and the principle of adaptation, the firing rate is reduced (say, to
highlights the properties that evoke optimal responses. It is ri-1). Upon increasing the stimulus to si+1, the firing rate
based upon the concept of selective response that we analyzed may return to ri, its initial value.”
above. Selectivity involves a non-differential mapping between
stimuli and responses, it grounds our current understanding of Garson is arguing that the relation between stimuli and firing
sensory cell types, and directly opposes Garson’s concept of rates is a many-to-many relation as opposed to a one-to-one
differential mapping. relation or a one-to-many relation. This sense of relation
But the fact that a scientific community makes use of feature corresponds to the use of relation in relational databases. For
selectivity rather than differential mapping to acquire evidence example, the relation between a person and that person’s driver’s
does not in itself deny differential mapping. We, too, are license number is one-to-one; the relation between each continent
questioning the community in its interpretations of evidence, and its countries is one-to-many—each continent can have
although we are not challenging its methodology. Garson argues multiple countries, but each country belongs to one continent;
that the differential mapping between stimuli and spike rates and the relation between academic articles and their authors is
logically follows from a conjunction of the principle of neuronal many-to-many—each article may have multiple authors, and
rate coding with the principle of adaptation: each author may contribute to multiple articles.
Garson believes that the many-to-many relation between
“While the principle of rate coding entails that the stimuli and spike rates supports a notion of arbitrariness, and that
frequency of the sequence of action potentials is an this arbitrariness, although not sufficient, is a necessary
exponential function of the magnitude of the stimulus, the constituent of a concept of information. But what is necessarily
principle of adaptation entails that upon application of a arbitrary about a many-to-many relation in contrast to a one-to-
constant stimulus, the frequency of the sequence of action one relation? A one-to-one relation, such as the driver’s license
potentials will diminish, and eventually such outputs will example above, may be as arbitrary—or more arbitrary—than a
stop being produced. Hence the relation between the many-to-many relation, depending upon how that relation came
sequence and stimulus is differential.” to be. Garson attempts to distinguish between arbitrary and non-
arbitrary causal relations by contrasting two neurophysiological
No further derivation is given, which is concerning since each mechanisms involved in auditory perception: the tympanic
principle taken individually contradicts differential mapping in membrane and the hair cells of the inner ear. He says that the
Garson’s sense. Rate coding is a form of non-differential relation between air pressure oscillations and the vibrations of
mapping: a specific stimulus intensity directly maps onto a the tympanic member is one-to-one, and is therefore non-
specific frequency of action potentials. How can Garson assume arbitrary; but that the hair cells of the inner ear undergo
non-differential mapping as a premise to establish differential adaptation, which implies a many-to-many relation between air
mapping? And the principle of adaptation is equally troubling; it pressure oscillations and hair cell firing, and is therefore
implies that a negative change in the frequency of action arbitrary. Garson believes that relations that come to be through
potentials maps onto a constant intensity of the stimulus. But mechanisms of adaptation and are many-to-many are arbitrary,
this directly conflicts with Garson’s claim that differences in and those that do not involve adaptation and are one-to-one are

7
non-arbitrary. But it seems obvious that one-to-one and many- action, or not seen during scanning at all. Researchers recorded
to-many relations may both be arbitrary or non-arbitrary. The correct and incorrect responses. The results of the identification
relevant difference that determines arbitrariness for Garson must test were statistically correlated with the fMRI data to identify
be the presence of adaptive mechanisms, but he uses neuronal spatiotemporal fMRI activity patterns that predicted correct
adaptation only as a means to justify the many-to-many relation responses on the identification test. Poppenk et al. speculated
between stimuli and responses. In this sense the argument is that some of the identified brain regions enact “processes
circular. associated with successful encoding of intentions (p911).”
This particular T/BR study is more complicated than the
3.2.3 Summary of ES/BR Studies ES/BR studies of Hubel-Wiesel from an interpretational
standpoint. Although T/BR studies are not necessarily more
Rigorous experimental neuroscience has demonstrated complicated than ES/BR studies, the complexity of Poppenk et
neurons that selectively respond to a wide range of measurable al.’s study is not atypical for fMRI studies that include cognitive
parameters across the five senses. Because selective responses tasks. Like S/A and ES/BR studies described above, the
reliably covary with stimulus properties, we can use empirical evidence in this T/BR study consists of selective
mathematical tools to predict stimulus properties given spike correlations, in this case between successful task completion and
trains, and predict spike trains given stimulus properties. The properties of brain areas. These selective correlations, like those
idea that there is a neural code, however, is a metaphor, since described in S/A and ES/BR studies, do not logically imply that
spike trains are not encoded with respect to the organism, the function of the identified brain regions is to perform a
anymore than English is encoded with respect to an English process directly related to the task. Poppenk et al. make no
speaker. Further, the idea that spikes trains carry information is attempt to understand the so-called processes of the identified
grounded in the experimental evidence of selective covariations, brain region other than to say that processes are associated with
but most commentators conflate or equate carrying information the task, but the observed form of this association is statistical
and selective covariation. correlation. Even if this association was selectively causal, we
still could not infer that the function of the brain region had to do
3.3 T/BR with completing the task, for the causal association could be
accidental. And even if the function of the brain region involves
Task/brain-response studies combine aspects of S/A and processes to complete the task, we do not know that completing
ES/BR studies, and our critiques of these studies will apply. In the task involves processing information of any kind.
addition to systematically manipulating the external environment
as in ES/BR studies, T/BR studies add to this manipulation a task 3.4 BM/BR
for the subject to perform. The task is similar to an ability in a
S/A study, except that the task is a transient activity while an Brain-manipulation/behavioral response studies demonstrate
ability is an ongoing capacity to act in a particular way. With the behavioral effects of causally manipulating brain properties.
regard to scientific research, tasks should have observable or As an example, optogenetic studies are a relatively recent
measurable criteria for successful completion, while abilities advance in BM/BR experimentation, and permit precise
should have observable or measurable criteria for possession of manipulation of neuronal activity. These experiments involve
the ability. We can often study a topic using either task or ability expressing light-sensitive genes within specific neurons or
language. For instance, in memory studies, we can assign populations of neurons in living animals. When the neurons with
subjects the task of memorizing a set or numbers, and then ask the expressed genes are exposed to light of a particular
for those numbers at a later time. The task of remembering and wavelength, the activity of the neuron will either increase or
the ability to remember are similar in that the criteria for decrease, allowing for precise control of the neuron’s activity.
completion of the task and possession of the ability are Presumably the expressed genes do not significantly alter the
equivalent. If one remembers the numbers correctly, one has functioning of the neuron otherwise. For example, Wyart et al.
successfully completed the task and possesses the ability to (2009) expressed light sensitive genes within so-called Kolmer-
remember. Agdur cells of the zebrafish, and then non-invasively
Let us recall the Poppenk et al. (2010) T/BR study on manipulated the neuronal activity of these cells which modulated
prospective memory described above, where subjects were the swimming behavior of the animal.
presented a series of visual scenes and instructed to either BM/BR studies establish causal relationships between the
imagine performing an action associated with that visual scene activity of multiple brain areas or between brain area activity and
(e.g. swinging on a swing when shown a swing), or to use the behavioral responses. Canonical examples of BM/BR
scene as a reminder to perform an action the next time the same experiments involve electrical stimulation of brain areas resulting
scene was viewed. FMRI was used to measure brain properties in muscle movements. These types of experiments can be traced
while the subjects performed these tasks. After this task, the back to at least Fritsch and Hitzig (1870) who applied surface
subjects were taken to a quite room to perform an identification electrodes to dog brain and demonstrated that the anatomical
test. They were shown visual scenes on a computer and asked to location of electrical stimulation selectively covaried with
indicate whether each scene was studied as an intention, an movements in different muscle groups. Neuroscientists name the

8
structure of this covariation a somatotopic organization, and transmission. Of course, one can use a metaphorical information
classically explain the relation between cerebral cortex and language to describe the propagation of electrical activity
muscle movements with the following three hypotheses involving the patellar reflex, but this language adds nothing to
(Graziano et al. 2002): (1) the precentral gyrus, or primary motor our physical understanding of the reflex. Electrical activity does
cortex, contains an explicit topographic map of the body with the not propagate randomly through the brain; its pathways are
foot on the top of the cerebral hemisphere, the mouth on the organized, and scientists attempt to understand this causal
bottom, and other parts systematically organized in between; (2) organization with respect to an organism’s abilities. BM/BR
the activity at each point in the map specifies the tension in a studies are an important tool for understanding this organization,
small group of muscle fibers; and (3) cortical motor areas are however, knowledge of causal organization does not imply
organized in a clear hierarchy of control. These classical knowledge of the processes that occur within a brain region, nor
hypotheses have been summarized as far back as 1938 by Fulton, does it provide evidence of information processing.
although there is a significant body of subsequent evidence that
is not compatible with the classical theory (Graziano et al. 2002).
Contemporay BM/BR studies of primate motor function 4. CONCLUSIONS
typically involve inserting microelectrodes into the cerebral
cortex of an awake animal, and injecting low electrical currents Cognitive and neuroscientists have measured action
into the cellular network while simultaneously recording the potentials, ischemic lesions, distributions of dopamine receptors,
pattern of muscle movements, allowing the researcher to catalog electromagnetic surface potentials, and many other brain
the causal relations between electrical stimulation and these properties. They have measured temporal sequences of brain
movements. One hopes or assumes that the patterns of muscle properties contemporaneous with environmental stimuli and
movements in response to exogenous stimulation are similar to behavioral actions, and have manipulated brain tissue and
those caused by endogenous neural activity, although the induced catalogued behavioral changes. It is clear that the activity of
electrical activity is clearly non-physiologic, complicating any particular brain areas selectively covaries with specific stimulus
interpretation of the results. properties and abilities. However, since we cannot logically
If we consider the brain to be a mechanical mechanism, what infer processes and functions given experimental correlations
do these motor BM/BR studies tell us about the brain? In other between stimuli or abilities and brain areas, it is difficult to
words, knowing that anatomical locations of electrical justify Bechtel and Richardson’s supposed uncontroversial claim
stimulation and patterns of muscle movement covary with each that “the brain contains some regions that are specialized for
other, what can we say about the processes or functions that processing specific types of information.” The brain does
occur in those brain areas? We might say that there is a causal contain regions that selectively covary with particular stimulus
propagation of electrical activity, beginning from motor cortical properties and particular abilities, and we may at times attribute
areas through the central nervous system and to spinal motor causality to these covariations, but interpretational claims that
nerves that enact patterns of muscle contracture. If we attribute information processing to brain areas go beyond the
electrically stimulate the brain in other areas, the electrical evidence.
activity does not propagate to spinal motor nerves. The motor We admit that our examination of the research is far from
cortex therefore acts as a metaphorical gateway or hub of exhaustive, and that we have limited our analysis to relational
electrical activity from CNS to spinal nerves, where the pathways evidence. There are likely other categories of evidence that
are at least partially organized with respect to specific muscle support information processing in the brain, but we feel we have
movements. examined evidence that is classically taken to support
Claims of motor information processing are not needed to information processing and find it inadequate.
describe these results, but presumably arise when one assumes Rather than inferring solely from the experimental evidence
that areas of the motor cortex naturally represent various muscle as we have suggested, cognitive scientists typically begin with a
groups, or that the precentral gyrus contains a map of the body, priori theories about particular cognitive processes, and use the
but the fact that electrical activity propagates through specific results of relational studies to choose between these theoretical
pathways does not establish that motor areas naturally represent cognitive processes (Henson 2006). One might call the claim
muscle groups. Just as the idea of a neural code is a sometimes that the brain processes information one of these hypothetical
useful metaphor for communicating scientific results, the idea of cognitive theories. We have not found empirical support for this
a topographic map of the body in the precentral gyrus is a useful cognitive theory, although others may interpret the evidence
metaphor for summarizing the data about causal organization otherwise. If one assumes that selective correlation or causation
with respect to electrical stimulation. implies information carrying, then one will see information
Compare the structure of the patellar reflex in the peripheral processing everywhere one looks. In this sense, information
nervous system: hitting the patellar tendon stretches the processing is not a scientific theory or fact, but a basic principle
quadriceps muscle which activates sensory receptors that or metaphor that many people find useful in communicating and
propagate electrical activity through motor neurons that contract interpreting evidence.
the quadriceps. We can describe the propagation of physical There is no doubt that mathematical modeling will help us to
changes without any reference to information processing or understand the functioning of the brain, but we have not

9
addressed modeling efforts in this work because modeling in Garson, J. (2003), The Introduction of Information into
cognitive science is underdetermined, and to our knowledge Neurobiology. Philosophy of Science, 70 (5): 926-936
there are no cognitive models that are largely accepted as
representing mental facts. Further, while modeling may provide Graziano, M.S.A., Taylor, C.S.R., Moore, T., and Cooke D.F.
some evidence for information processing in the brain, this (2002), The Cortical Control of Movement Revisited. Neuron,
evidence is relatively weak because of the above and because 36: 1-20.
information processing enters as an assumption of these models
and not as a consequence. Hanson, Norwood R. (1958), Patterns of Discovery. Cambridge:
University of Cambridge Press.

Hardcastle, V. G. and Stewart, C. M. (2002), What do brain data

5. REFERENCES really show?, Philosophy of Science, 69(3), S72-S82

Adrian, E.D. and Y. Zotterman (1926), The Impulses Produced Henson, R. (2006), Forward Inference Using Functional
by Sensory Nerve Endings: Part II: The Response of a Single Neuroimaging: Dissociations Versus Associations. Trends in
End Organ, J. Physiol. Lond. 61: 151-171. Cognitive Science, (10) 2, 64-69.

Adrian, E.D. and R. Matthews (1927), The Action of Light on Hodgkin A and Huxley A. (1952), A Quantitative Description of
the Eye. Part I, J. Physiol. Lond. 63: 378-414. Membrane Current and its Application to Conduction and
Excitation in Nerve. J Physiol, 117:500–544.
Bechtel, W. (2008). Mental Mechanisms: Philosophical
Perspectives on Cognitive Neuroscience. London: Routledge. Hubel, D.H. and Wiesel, T. N. (1962), Receptive Fields,
Binocular Interaction and Functional Architecture in the Cat's
Bechtel, W. and Richardson, R. C. (2010). Neuroimaging as a Visual Cortex. J. Physiol., 160, 106-154,
tool for functionally decomposing cognitive processes. In S. J.
Hanson and M. Bunzl, Foundational issues in human brain Hubel, D. H. (1962), Transformation of Information in the cat's
mapping. Cambridge, MA: MIT Press. visual system. Proceedings of the International Union of
Physiological Sciences, 3: 160-169.
Brewer, W.F. and B.L. Lambert (2001). The Theory-Ladenness
of Observation and the Theory-Ladenness of the Rest of the Klein, Colin. (2010), Images Are Not the Evidence in
Scientific Process, Philosophy of Science, 68(3): S176-S186. Neuroimaging. The British Journal for the Philosophy of
Science, (61)2: 265-278.
Broca P (1861) Nouvelle observation d'aphémie produite par une
lésion de la moitié postérieure des deuxième et troisième Kuhn, Thomas. S. (1962), The Structure of Scientific
circonvolution frontales gauches. Bull Soc Anat, 36:398-407. Revolutions. Chicago: University of Chicago Press.

deCharms R.C. and Zador A. (2000) Neural Representations and Logothetis, Nikos K., and Brain A. Wandell (2004), Interpreting
the Cortical Code. Ann. Rev. Neurosci., 23:613-647. the BOLD Signal, Annual Review of Physiology, 66: 735-769.

Dronkers, N.F. (2000) The Pursuit of Brain-language Logothetis, Nikos K. (2008), What We Can Do and What We
Relationships. Brain and Language, 71(1):59-61. Cannot Do with fMRI, Nature, 453: 869‐878.

Floridi, Luciano, (Winter 2008 Edition) Semantic Conceptions of Millikan, R.G. (1989), Biosemantics. The Journal of Philosophy,
Information, The Stanford Encyclopedia of Philosophy, Edward 86(6) 281-297.
N. Zalta (ed.), URL =
<http://plato.stanford.edu/archives/win2008/entries/information‐s Poldrack, R. A. (2006), Can Cognitive Processes be Inferred
emantic/>. from Neuroimaging Data? Trends in Cognitive Sciences, 10(2),
59-63.
Fritsch, G., and Hitzig, E. (1870), Ueber die elektrishe
Erregarkeit des Grosshirns, translated by G. von Bonin, The Poppenk J., Moscovitch M., McIntosh A.R., Ozcelik E., Craik
Cerebral Cortex, W.W. Nowinski, ed. (Springfield, IL: Thomas, F.I.M. (2010), Encoding the Future: Successful Processing of
1960), pp. 73–96. Intentions Engages Predictive Brain Networks. Neuroimage, 49,
p905-913.
Fulton, J.F. (1938), Physiology of the Nervous System. New
York: Oxford University Press. Popper, K.R. (1959), The Logic of Scientific Discovery, London:
Hutchinson.

10
Rieke, Fred, David Warland, Rob de Ruyter van Steveninck, and
William Bialek (1999), Spikes: Exploring the Neural Code.
Cambridge, MA: MIT Press.

Roskies, Adina L. (2007), Are Neuroimages Like Photographs of

the Brain?, Philosophy of Science, 74: 860-872.

Shannon, Claude E. (1948), A Mathematical Theory of

Communication, The Bell System Technical Journal, 27:
379‐423, 623‐656.

Uttal, W. R. (2001). The new phrenology: The limits of localizing

cognitive processes in the brain. Cambridge, MA: MIT Press.

Wyart C, Del Bene F, Warp E, Scott EK, Trauner D, Baier H,

Isacoff EY. (2009), Optogenetic dissection of a behavioural
module in the vertebrate spinal cord. Nature. Sep
17;461(7262):407-10.

Yerkes, R. M., & Dodson, J. D. (1908), The Relation of Strength

of Stimulus to Rapidity of Habit-formation. Journal of
Comparative Neurology and Psychology, 18, 459-482.

Zhang F, Aravanis AM, Adamantidis A, de Lecea L, Deisseroth

K (2007). Circuit-breakers: Optical Technologies for Probing
Neural Signals and Systems. Nat. Rev. Neurosci., 8(8): 577–81

11