Sensory and Perceptual Capacities

unclesamnorweiganIA et Robotique

18 oct. 2013 (il y a 5 années et 2 mois)

137 vue(s)


Sensory and Perceptual Capacities

Philosophers have devoted considerable energy to addressing the question of how we ought to
distinguish the various sensory modalities from one another. Meanwhile, relatively little
attention has been paid to the prior question of what having a sense or a
sensory capacity
amounts to. Further, this familiar debate on individuating sensory modalities has from the outset
been explicitly concerned with the project of distinguishing types of
, while very little
is said in this context about what
capacities are and how they relate to sensory ones.
In what follows we hope to make progress on these foundational issues.

1. Sensory Capacities

The concept of a sense figures in everyday thinking and in various disciplines. We assume that a
sense i
s a capacity of a

system. We speak of the capacities of
rather than

because inorganic artefacts can and do possess sensory capacities. (For example, machine vision
is a discipline devoted to constructing machines that see.) On our usage,
might be a
system proper, a subsystem, or a supersystem. There is no
a priori

reason to deny that sensory
capacities can belong to subsystems and supersystems as well as systems proper.

When does a system have a sense? What conditions must be met?

We begin with a
promising candidate to serve as a necessary condition on the presence of a sense:

Behavior Constraint. A system has a sensory capacity only if it is capable of discriminatory


Burge (2010: 377) has this constraint in mind when h
e writes: “The notion of a sensory system is
a functional notion. A sensory system is a system of an entity capable of behavior.” Likewise,
Keeley (2002: 15) expresses sympathy with the Behavior Constraint in the following remark on
sensory modalities: “Pa
rt of what it means to have a modality is to be able to make behavioral
discriminations within that modality.” We follow Keeley and Dretske (1988) in taking
to be
endogenously produced movement

A straightforward problem with the Behavior Constrai
nt is that it is too narrow: sensory
capacities can exist in the absence of a capacity for movement. Thanks to brain
interfaces, even people suffering complete paralysis can engage in back
communication with others. For example, they can

the use of an fMRI by
performing various mental tasks each of which is mapped to a specific letter of the alphabet
(Sorger, Reithler, Dahmen, & Goebel 2012). This method of communication involves
responding to questions that are seen or he
ard and so involves the exercise of a sensory capacity
in the complete absence of any capacity for movement.

In light of this problem with the Behavior Constraint, one might suggest the following
alternative constraint on sensory capacities:

ion Constraint. A system has a sensory capacity only if it is capable of
discriminatory behavior or discriminatory mental action.

The Behavior
Action Constraint straightforwardly accommodates our paralysis case, but we
worry that even after this emendatio
n we still have something too narrow. Consider the
regulation of circadian rhythm by the brightness of light registered at the retina. Here the
response is neither bodily movement nor mental action, and yet there is still photoreception and

an endogenously

generated response, just as in standard cases of visually guided discriminatory
behavior/action. It seems arbitrary not to count these photoreceptive reactions as exercises of a
sensory capacity.

The need for something like the Behavior Constraint is evid
ent. It is a conceptual truth
that a sensory system is capable of output responsive to differences in the sensory input. A sense
is a discriminatory capacity and having this sort of capacity for discrimination goes beyond
having the capacity to carry infor
mation about a stimulus type. Under certain conditions, a table
will be differentially affected (with respect to its temperature) by differences in the intensity of
ambient light and will thereby carry information about those differences. Nonetheless, in t
usual case a table is not a system with a sensory capacity. A system with a sensory capacity has
to have enough internal complexity to warrant a distinction between a signal transduction phase,
an encoding

receptors of information from the environme
ntal stimulus, and an output phase,
the system’s discriminatory response to the sensory signal.

The following is our attempt to articulate a suitably broad account of sensory capacities,
one that can accommodate discriminatory output which takes forms othe
r than movement:

A system has a sensory capacity if and only if (i) the system can encode information from a type
of environmental stimulus

transducing receptor(s), (ii) the system is capable of
endogenously produced discriminatory response(s) to tha
t stimulus type, and (iii) the system’s
encoding phase is sufficiently distinct

from its output phase.


We are using the phrase ‘environmental stimulus’

in a broad way to allow that some stimuli may be part
of a creature’s internal environment (e.g. noxious stimuli).


We shall not attempt to specify what, generally speaking, makes an encoding phase
sufficiently distinct

from a response phase. It is doubt
ful that there is any useful criterion that would apply to all cases.


For the remainder of this section we explore some potential objections to this proposal.

There are at least two types of objection one might raise to
our account of sensory
capacities. First, one might argue that our account involves an unmotivated departure from our
everyday or traditional notion of a sense. Second, one might worry that our account strays too far
from the notion of sensory capacities a
t work in actual scientific practice. We take both of these
kinds of objection seriously. We begin by addressing the first.

Our everyday notion of a sense is plausibly regarded as a component of folk psychology.
One might argue that processes which are al
together non
mental would not count as sensory, on
our usual way of thinking about the sensory. For example, the photoreceptive processes involved
in the regulation of the size of the pupils would not count as sensory. The same goes for the
processes which serve to monitor carbon dioxide levels of the blood.

account seems to have the consequence that these processes count as sensory, and so our account
fails to capture our everyday notion of a sense. Of course, this sort of departure fro
m tradition is
not by itself problematic. The deeper worry is that we have offered no reasons for preferring a
revisionary approach to the sensory.

This objection rests on a misunderstanding of our everyday notion of a sense, which is as
much a part of fo
lk biology as folk psychology. Biologists are not abandoning our everyday
notion of a sense when they extend it to creatures whose behavior is not susceptible to (folk)
psychological explanation, as is common, e.g., in botany. Indeed, in English we have a
term for sensory states that have a decidedly psychological character, namely, ‘perception,’ a



(forthcoming) makes claims of this sort about the pupillary light reflex and carbon dioxide
monitoring, emphasizing that these processes do not fall under “the traditional notion of a sense.”


term which has its origins in folk psychology. We turn to the topic of perception in the following

As it turns out, there is good precedent wi
thin the sciences for treating the pupillary light
reflex as driven by sensory inputs.

The same goes for the photoentrainment of circadian rhythm
and the chemoreceptive processes that regulate carbon dioxide levels in the blood. Nevertheless,
one might th
ink that there is something unnecessarily abstruse about extending the notion of a
sense to suborganismal items like the respiratory system. Why not avoid this awkwardness by
endorsing Keeley’s seemingly truistic claim that a sense is “an avenue into an

We resist the temptation to make reference to whole organisms in our account on the
grounds that it would be arbitrary to place this limitation on sensory capacities

and not just
because there are perfectly good examples of sensory capacities in inorganic
systems. A
unicellular paramecium exhibits discriminatory behavior. Notice, though, that a sperm is capable
of the same sort of behavior

i.e. chemotaxis and thermotaxis

so the chemoreception and
thermoreception that guide the sperm’s behavior should count
as sensory capacities just as much
as the chemoreception and thermoreception present in the paramecium. But a sperm is not itself
an organism; it is a suborganismal entity. Similarly, we might attribute behavior to a colony of
organisms that itself looks a
nd behaves like an organism (e.g. the Portuguese man o
war, which
looks and stings like a jellyfish). Since it can exhibit the type of behavior characteristic of whole
organisms, we have no principled reason to deny that a colony of this sort can possess s
capacities. The general point here is that behavioral kinds like chemotaxis and thermotaxis cut


Gamlin (2004) writes, for example, that “
The human pupillary
light reflex is one of the most familiar
and well studied circuits that link human vision directly to motor behavior.” If the pupillary light reflex is
an aspect of vision, of course, it is a genuinely sensory phenomenon. This outlook is reflected in the
ignificant number of papers published on the pupillary light reflex in journals like
Journal of Vision
Vision Research


across boundaries like organism vs. subsystem or supersystem. Likewise, the same kinds of
sensory processes (e.g. chemoreception and thermoreception) are

present whether we are dealing
with paramecia, on the one hand, or sperm and human subsystems, on the other.

The next worry about our account of sensory capacities is that it is too close to our
commonsense view, and consequently not suitable for the purp
oses of scientists working in
disciplines like sensory biology and neuroethology. Our everyday notion of a sense extends to
capacities with miraculous or even magical origins. Keeley suggests that we get something closer
to the kind investigated by scienti
sts if we add the following constraint:

Neurobiological Constraint. A system has a sensory capacity only if its encoding phase arises
evolutionarily dedicated sensory neurons.

Keeley’s constraint has the advantage (from our point of view) that it acc
ommodates the cases
discussed above. The pupillary light reflex and the photoentrainment of circadian rhythm both
involve retinal neurons dedicated to extracting information from light. And the regulation of
respiration depends on neurons of the medulla

e “central” chemoreceptors

dedicated to
extracting information from the chemical concentrations of the blood.

A significant problem with Keeley’s Neurobiological Constraint is that it arbitrarily
excludes the paramecium from the class of entities which po
ssess chemical senses. The single
celled paramecium has no neurons and hence cannot satisfy the Neurobiological Constraint. But
the use of specialized nerve cells is just

one way

that an organism can sense its environment. A
more rudimentary way is

cialized receptors on the surface of an all
purpose cell. And this
latter route is taken by the paramecium. In the words of neuroscientist Ralph J. Greenspan:


Senses go back a long way... Chemical sensing is almost certainly the original sense,
given that

life arose in the liquid environment of the sea and that even bacteria have a
neuronal version of it. But with the arrival of multicellular animals, separate sense
organs arose and with them the ability to see and hear as well as taste, smell, and tou
(2007: 2)

Sensory capacities exist in the absence of distinct sense organs, and even in the absence of
neurons altogether.

Keeley might respond that even in the case of the paramecium something similar to a
neuron is present, and hence the spirit of th
e Neurobiological Constraint is not genuinely
threatened. After all, paramecia (as well as human peripheral chemoreceptors) have membranes
with electrical properties like those of neurons (L
Barneo et al. 2008; Jegla & Salkoff 1995).
The problem, howev
er, is that a sense can also be present in a system with a very different
constitution. Some robots can see, despite the fact that they do not contain cells with electrically
excitable membranes. There is little temptation, then, to adopt something like th
Neurobiological Constraint.

We turn, finally, to the most serious problem facing our approach to sensory capacities.
Our account seems to have the consequence that the average somatic cell possesses sensory
capacities. Cells of this sort routinely detect

and respond to chemicals in their local environment.
A skin cell in your foot, for example, might respond to chemical signals from neighboring cells
by increasing the expression of a certain gene, and thereby satisfy our criteria for possession of a
ry capacity. No doubt some will find this consequence of our account unacceptable. The
problem is not just that it is counterintuitive to include these cells among systems with sensory

capacities; the deeper worry is that scientists studying the chemical s
enses do not typically regard
the intercellular signaling of somatic cells to be among their objects of study.

Although our account appears to be at odds with the notion of a sense at work in actual
scientific practice, arguably there is something arbitrar
y in the way that sensory biology
demarcates the lower boundaries of the sensory. After all, some somatic cells manifest the same
capacities present in organisms like the paramecium. Both brain cells and white blood cells
exhibit chemotaxis, so they are ev
idently capable of chemoreception just as much as the
paramecium is.

And while many somatic cells do not display chemotaxis, they do detect and
respond to chemicals in their environment, thereby exemplifying a primitive form of

We would pr
efer not to stray needlessly from standard usage of the terms ‘sense’ and
‘sensory’ in everyday life and scientific practice, but our account of sensory capacities is not an
attempt to capture the nominal essence of these terms. Rather, our goal has been t
o indicate some
significant continuities
in rerum natura
, from visual processing in our heads to intercellular
signaling in our toes. We turn next to a significant discontinuity or division within the class of
sensory capacities, namely, the division betwe
en perceptual and non
perceptual sensory


See Rao et al. 2002 for a comparison of the chemotaxis of brain cells with that of white blood cells
(“leukocytes”). While the aut
hors reserve the term “chemotaxis” for white blood cells and speak only of
neuronal “migration,” they note that “
the guidance cues and receptors for neurons and leukocytes are used
in both systems, supporting a conservation of guidance mechanisms for cells

of distinct types.” Both
neurons and white blood cells utilize chemical cues to guide their movement.


2. Perceptual Capacities

When should we say that two creatures possess the same general type of sensory modality (e.g.
vision)? That is, what criteria should we rely on in making judgments of samenes
s or difference
with respect to sense modalities? Philosophical debate over this issue has been specifically
concerned with individuating
capacities understood as psychological or mental
abilities. Keeley (2002) and Ross (2008) are exemplary in
this respect. Both are explicit that they
are interested in modes of perceiving and that the modalities in question (seeing, hearing, etc.)
are psychological/mental kinds. For example, Keeley tells us that each form of perception “is a
potential mode by wh
ich information in the environment can pass through some boundary and
enter into the psychological system” (12). On this way of thinking about the philosophical issue,
criteria for the presence of, say, vision in a system will have to include some test for

whether the sensory inputs are feeding into a genuinely psychological system.

Keeley fails to engage sufficiently with this aspect of the project. Keeley’s
Neurobiological Constraint is concerned with the encoding or input phase, and does not

significant restrictions on what type of system the sensory input is feeding into. It is more natural
to look to a system’s output in determining whether a system is genuinely psychological in
nature, but Keeley offers nothing more than the Behavior

Constraint. As Ross notes, this
constraint is too weak: non
psychological systems could meet this and Keeley’s other constraints
on perception. Consider the photoreceptive processes that control the pupillary light reflex. This
reflex behavior can be pres
ent even in cases of complete cortical blindness, cases where the
relevant retinal activations clearly do not feed into a psychological system.


Keeley might resist the idea that pupillary dilation and restriction count as behavior (endogenously
produced movement), but there are other cases invo
lving more obvious instances of movement. For

Ross himself does not provide a comprehensive account of perceptual capacities. His
focus is on the more specif
ic issue of whether the human vomeronasal system counts as
genuinely perceptual, and so different from the non
mental, physiological processes driving the
pupillary light reflex. Ross’s suggestion is that a system has a form of perception only if it is
able of discriminations of a qualitative determinable (e.g color, sound, or flavor). He appeals
to psychophysical techniques of multidimensional scaling in an attempt to make his suggestion
more precise.

We reject Ross’s suggestion on the grounds that it
rules out the very possibility of simple
off” perceptual systems. For example, a pinhole visual system which detects nothing more
than the presence or absence of light at the receptor could not qualify as perceptual because the
system would not afford
discriminations within a qualitative determinable. We should leave open
the possibility that a pinhole visual system might feed into a genuine psychological system, and
thereby afford perception.

In any case, an appeal to qualitative determinables by its
elf won’t take us very far. An
account of perceptual capacities needs to have the result that sensory systems like the following
lack perceptual capacities:

example, the chemoreceptive processes that influence breathing rate in humans do not make sensory
information available to the subject, and yet they regulate the visible movements involved in breathing.


dimensional scaling “is a set of statistical techniques which can generate a spatial representation of
the relative qualitative similarities among a range of qualitative properties such as colors” (Ross 2008).
As Ross notes, the most familiar representatio
n of this type is “the psychological color space, with
dimensions of hue, saturation, and lightness.”


Even on the supposition that mentality requires phenomenal consciousness

a supposition we reject

there is no reason to think that a pinhole visual syste
m cannot feed into a genuine psychological system.
For there is no reason to think that a pinhole visual system cannot yield phenomenally conscious sensory


Some invertebrates display forms of behavior that are wavelength
selective in the sense
that the or
ganism responds differently to stimuli based on their spectral power
distribution and independently of the relative intensity of the stimuli. These organisms
would seem to satisfy the criterion for the possession of color vision. Their
discriminatory behav
ior may be the result, however, of their having different receptor
types independently driving different response mechanisms. A receptor most sensitive to
long wavelengths may be connected to a system that causes the organism to move
towards the light sour
ce and a receptor most sensitive to short wavelengths may be
connected to a motor system that causes the organism to move away from the light
source. (Hilbert 1992: 357)

Systems of this sort evidently satisfy our requirements on possession of sensory capac
ities. They
possess color vision in the most minimal sense that they are differentially responsive to
differences in wavelength. Notice, though, that any talk of color
is out of place.
Perception is a psychological capacity. The sensory inputs i
n a perceptual system feed into a
psychological system, and a psychological system must possess some degree of flexibility in its
responses to those sensory inputs. Flexibility is, in part, what sets psychological systems apart
from automata, the human res
piratory system, and spermatozoa. The invertebrates described by
Hilbert above are capable of discriminatory responses to stimuli, but their responses are mere
, not
. Their behavioral outputs are passively controlled by the stimuli. Because
they lack flexibility in their discriminatory responses, these systems are mere
, not

This example due to Hilbert illustrates the point that a system

capable of perception must
in its discriminatory responses to stimuli. Flexibility, however, is not always

psychological in character. Consider flexibility in the form of habituation. If an organism is
repeatedly exposed to a given stimulus ty
pe, the magnitude of the organism’s response to stimuli
of that type may decrease. The sea slug
Aplysia californica

withdraws its gill and siphon when
its siphon is tactually or electrically stimulated. The amplitude of the withdrawal reflex decreases

repeated stimulation. Sensitization serves as another example of minimal flexibility. In
cases of sensitization the magnitude of a response

with successive presentations of a
stimulus. We should not regard the presence of these forms of flexibil
ity in a system as sufficient
for it to count as a psychological system. All animals and many other organisms seem capable of
habituation or sensitization (Burge 2010: 306 n25). The notion of a psychological system does
not apply so widely.

Matthen (2005,

forthcoming) suggests that a system has the sort of flexibility required for
perception if it is capable of associative learning, e.g. if its discriminatory responses are subject
to conditioning. In
conditioning a response is transferred from on
e stimulus type to
another. Take the salivation response which the presentation of treats naturally elicits in domestic
cats. It is easy to transfer this sort of response to a different type of stimulus. For example, if one
reliably precedes the presentati
on of treats with a distinctive type of noise, like that caused by
shaking the container for the treats, the salivation response will eventually extend to this
previously neutral stimulus. The sound caused by shaking the container will, by itself, elicit t
response in question. In
conditioning responses are modified through reward or
punishment. For example, domestic cats can be trained, using treats as rewards, to stand briefly
on their hind legs in response to a cue like the utterance of a parti
cular phrase. Operant
conditioning is present if the reward brings about a modification of the cat’s behavior, increasing
its tendency to stand on its hind legs in response to the cue. Whether we are dealing with

classical or operant conditioning, Matthen
thinks that a system capable of associative learning
has the sort of flexibility required to count as a perceiver.

The kind of flexibility Matthen has identified is not quite what we are looking for. A
perceptual system, we assume, is capable of
of classification or sorting. That is, a perceptual
system must be capable of
in its discriminatory responses to stimuli

even if only a very
primitive sort of agency. Agency requires more than just flexibility in how a system responds to
sensory inp
uts; it requires flexibility in the pursuit of the system’s contingent needs or desires.
Accordingly, being subject to classical conditioning is not sufficient for the presence of agency.
Classical conditioning involves the flexible
of a respo
nse from one stimulus type to
another, but transference of a response can occur independently of needs or desires. In this
respect classical conditioning differs importantly from operant conditioning. In paradigm
instances of operant conditioning we find p
ursuit and avoidance behaviors flexibly guided by
recognitional capacities. Take the familiar case of operant conditioning in honeybees. Suppose
we expose bees to an array of receptacles differing only in color and whether they contain nectar.
Suppose furt
her that the bees consistently find the desired nectar only in blue containers. When
confronted with similar displays on subsequent occasions, hungry bees will fly directly to the
blue receptacles. In preferring the blue containers, the bees are evidently
exercising rudimentary
recognitional capacities. They recognize the blue ones and fly directly towards them. This
flexible response to the relevant stimulus is straightforwardly purposive in character: it is
contingent on the need or desire for food.

The f
lexibility involved in classical conditioning extends to systems that appear to lack
psychological agency, systems like the guinea pig’s immune system. In an early study of
immune system learning related by Pacheco
pez et al.

, guinea pigs were exp
osed daily

to scratching or heating of the skin prior to being injected with a foreign substance. The injection
resulted in an increase in white blood cell count. Eventually the immune response occurred (in an
attenuated form) when the skin was scratched o
r heated in the absence of an injection. The
immune system transferred an unconditioned response to a stimulus (increase in white blood
cells) to a previously neutral stimulus (scratching or heating). Classical conditioning is evidently
present, and yet we

should hesitate to think of the immune system as exhibiting the sort of
flexible agency present in honeybees. Learned responses in bees are not only contingent on the
presence of the relevant stimulus; they also depend on current needs or desires. The res
ponse of
the guinea pig’s immune system is not similarly contingent on needs or desires. Agency is
evidently absent.

These worries about Matthen’s proposal suggest that operant conditioning alone is key to
understanding what is distinctive about perceptual

capacities. Our preferred account of
perceptual capacities can be summed up as follows:

A system has a perceptual capacity if and only if (i) the system can encode information from a
type of environmental stimulus
transducing receptor(s), and (ii) th
e system is subject to
operant conditioning with respect to that stimulus type.

The remainder of the paper is devoted to addressing some objections to this way of thinking
about perception.


Note that operant conditioning, as we are understanding it, is possible in the absence of
a capacity for
movement. For example, the paralyzed subjects discussed above are all, in principle, susceptible to
operant conditioning. Their mental actions might be modified by reward or punishment.

Note also that we are not attempting to give a reductiv
e definition of perception in non
psychological terms. We allow that perhaps operant conditioning must be understood by reference to an
organism’s desires, and that desire might itself be an irreducibly psychological notion.


Some might object that, on our account, perceptual capacities ca
n exist in systems which
lack phenomenal consciousness. This consequence is, however, by design. As Burge notes
(2010: 374
376), perceptual psychology recognizes a variety of states as perceptual that are or
might be non
conscious. Examples include blindsi
ght and other neurological syndromes,
perception in organisms such as bees and spiders, and the early stages of human visual
processing. Our interest is in the psychological notion of
, not the philosophical notion
perceptual experience
. Only
the latter necessarily involves phenomenal consciousness.

A second worry is that our account involves a view of psychological agency that is too
permissive. The worry is that our account does nothing to ensure that perceivers possess a
desire psych
. Susceptibility to operant conditioning plausibly requires desires and
recognitional capacities, but it does not obviously require beliefs. Some philosophers will reject
out of hand any view of agency that omits a fundamental role for belief (see Car
ruthers 2005:

Why draw a distinction between sensory recognition (e.g. seeing as) and perceptual
belief? A simple example will illustrate the need for a distinction along these lines. Suppose you
are faced with what you know to be an illumination ed
ge in the scene before your eyes. In many
cases of this sort it will be possible for you to make a Gestalt shift and see the edge as a
reflectance edge. Seeing a luminance edge as a reflectance edge (rather than as a difference in
illumination) is distinct

from having the correlate perceptual belief that the edge is a reflectance
edge (and not an illumination edge). Perhaps both involve noticing (visual attention) and

recognizing (seeing as), but only perceptual belief essentially involves
. That
is, only belief essentially involves
regarding as true

Belief is in each case a creature’s point of view on how things are, a point of view from
which certain actions make sense and others don’t, a point of view that we can appeal to in
tion of behavior. Belief
desire explanations of behavior make action intelligible by
clarifying why the action

makes sense

from the subject’s point of view. Perceptual belief is
essential here because it captures the viewpoint of the subject on how things
are: perceptual
belief constitutes a subject’s take on
how things are

and not just
how things look or appear
Sensory recognition need not capture the subject’s take on how things are, as our reflectance
edge example illustrates. Accordingly, we should hes
itate to think that sensory recognition can
serve as a stand
in for perceptual belief. Psychological agency may well require the latter.

This line of reasoning is persuasive when we are dealing with creatures that draw a
distinction between appearance and
reality, creatures with the concept of truth. In sophisticated
creatures like us there is an important distinction to be drawn between sensory recognition and
perceptual belief. Only the latter necessarily captures our point of view on how things are. The
situation is rather different, however, when we turn to creatures that lack metacognitive
distinctions and concepts. Arguably, the distinction between sensory recognition and perceptual
belief collapses where these sorts of creatures are concerned. A honey
bee might be responsive to
how things are without being able to contrast how things are with how things seem, without


This distinction between seei
ng as and believing could be manifest in behavior. We could give subjects
the following instructions: press button 1 if you succeed in seeing the difference in the display as a
reflectance edge; then press 2 if you believe that it is. (Since belief comes i
n degrees, we might prefer to
measure the subjects’ degree of confidence.) The distinction between seeing as and believing could also
be manifest in matters of normative assessment. Suppose that it is epistemically irrational to believe that
an edge is a r
eflectance edge. We needn’t think that it is epistemically irrational to see the edge as a
reflectance edge.


having the concept of reality as something different from appearance. Having the capacity to
recognize how things are might afford the hon
eybee a point of view. From that point of view,
some actions might make sense, others not so much, and the bee acts accordingly. Some of the
bee’s behavior is now best explained by appeal to this point of view. That is, the core elements of
desire p
sychology are present.

Our first response, then, is that our view of agency is not as revisionary as it might first
appear. The second point we wish to emphasize is that we, like Keeley, are attempting to
articulate an account of perceptual capacities that

can serve the purposes of perceptual
psychologists. With this end in mind we are particularly keen on having our account extend to
the birds and the bees.

We grant that there are other contexts in which a more stringent criterion
for agency is desirable.

A third and final point to keep in mind is that those unhappy with our criterion for agency
might still be in broad agreement with us on a more fundamental issue, namely, the issue of how
to distinguish perceptual capacities from mere sensory ones. On thi
s matter we part ways with
Burge (2010), who sees the distinction as a difference at the stage of encoding rather than a
difference on the output side. On Burge’s account, perceptual states are

that attribute properties to the enviro
nment and hence can literally be accurate or inaccurate.
States that are merely sensory, by contrast, may reliably covary with the environment, but they
do not literally represent anything. Burge’s account commits him to rejecting the idea that
flexible ag
ency distinguishes perceptual capacities from merely sensory ones. Indeed, he remarks


Carruthers (2005) argues that honeybee behavior is best accounted for on the assumption that bees are
capable of spatial reasoning from beliefs a
nd desires. For a more deflated interpretation of insect
navigation skills generally, see Burge (2010: 492
518). Since there are legitimate grounds for doubt about
whether bees have beliefs, we prefer an account of perception that does not presuppose the c
apacity to
form beliefs.


that a “perceptual system could be as innately constituted and as hard
wired as one pleases and
still engage in perception” (307).

Elsewhere we have responded at length
to Burge’s defense of the idea that perception
involves a special type of sensory encoding, namely, representing [reference omitted]. We do not
believe it succeeds. Our present account can be seen as one way of fleshing out an alternative
idea: perceptual
capacities are distinguished from mere sensory ones by a special type of output,
not a special type of encoding. Those who believe our account is on the right track but disagree
with our particular criterion for agency (i.e. susceptibility to operant condi
tioning) can adjust our
account accordingly.


[One reference has been omitted for blind review]

Burge T. (2010).
Origins of objectivity
. New York: Oxford University Press.

Carruthers P. (2005).
Consciousness: Essays from a higher
order pers
. New York:
Oxford University Press.


Burge makes this comment in the course of rejecting the idea that perception requires a capacity for
. But if Burge rejects

requirement, then he must also reject the idea that some more stringent
criterion of agency
distinguishes the perceptual from the merely sensory.


We allow that systems with perceptual capacities may also need representational capacities
possessors of psychological agency. But in that case it is the recognitional states or belief states that

need to be representational in character. There is no obvious need for the sensory states themselves to be
representational, on our account.


Dretske F. (1988).
Explaining behavior: Reasons in a world of causes
. Cambridge, MA: MIT

Gamlin P. (2004). Primate pupillary responses mediated by a novel photopigment.
Journal of

4, 11. doi:

Greenspan R.J. (2007).
An introduction to nervous systems
. Cold Springs Harbor: Cold Spring
Harbor Laboratory Press.

Hilbert D. (1992). What is color vision?
Philosophical Studies

68: 351

Jegla T. & Salkoff L. (1995). A multigene

family of novel K+ channels from
Receptors and Channels
3 (1): 51

Keeley B. (2002). Making sense of the senses: Individuating modalities in humans and other
Journal of Philosophy

99 (1): 5

Barneo J., Ortega
Sáenz P., Pardal R., Pascual A. & Piruat J.I. (2008). Carotid body
oxygen sensing.
European Respiratory Journal

32: 1386

Matthen M. (2005).
Seeing, doing, and knowing: A philosophical theory of sense perception.
New York: Oxford University Press.

Matthen M. (forthcoming). The individuation of the senses. In M. Matthen (ed.),
handbook of the philosophy of perception
. Oxford University Press.

López P., Niemi M.B., Engler H. & Schedlowski M. (2007). Neuro
immune associative

In F. Bermúdez
Rattoni (ed.),
Neural plasticity and memory: From genes to brain
. Boca Raton: CRC Press.

Rao Y., Wong K., Ward M., Jurgensen C. & Wu J.Y. (2002). Neuronal migration and molecular
conservation with leukocyte chemotaxis.
Genes & Deve
16: 2973


Ross P. (2008). Common sense about qualities and senses.
Philosophical Studies

138 (3): 299

Sorger B., Reithler J., Dahmen B. & Goebel R. (2012). A real
time fMRI
based spelling device
immediately enabling robust motor
dent communication.
Current Biology

22 (14): 1333