<< . .

( 29)

. . >>

required. What do we need a theory of natural information for? In this
context, we require it to support a theory of “intentional” representa-
tion, in the sense introduced by Brentano. This is the kind of represen-
tation that displays Brentano™s mark of the mental. Intentional represen-
tations can represent nonexistent things, for example, nonexistent facts.
They can be misrepresentations. All agree, of course, that natural infor-
mation is not itself intentional, that it cannot misrepresent or be false.
“Informational semantics,” as Fodor calls it, is an attempt to show how,
despite this difference, intentional representation still rests at base on
natural information.
How to move from a theory of natural information to a theory of
intentional representation is, however, a problem. That is what Fodor™s
theory of “asymmetrical dependency” is designed to do (1990, Chapter
4). And that is what Dretske™s addition of teleology is designed to do “
his claim that it is only a function, not always realized, of intentional
representations to carry natural information (1981, 1991). Fodor™s asym-
metrical dependency theory seems, quite explicitly, to rest on informa-
tionL, but I won™t argue that case here. Rather, I will try to show how
teleology can be combined with a theory of soft natural information to

produce the variety in forms of intentional representation that animals
require. But there has been some confusion about the relation of teleo-
logical accounts of intentionality to informational semantics. So let me
first remark on that relation.
Naturalized teleological theories of the content of representations are
attempts to explain Brentano™s mark of intentionality: How can repre-
sentations be false or represent nonexistent things? But teleological the-
ories are only overlays, minor additions, veneers superimposed, on prior
underlying theories of representation, and there can be considerable va-
riety among these underlying theories. When looking at any teleologi-
cal theory, the first thing to ask is on what kind of more basic theory of
representation it rests.
Suppose, for example, that you think of mental representations as
items defined in a classical functionalist way, in accordance with pat-
terns of causal/inferential dispositions. And suppose that you have a
theory that tells what dispositional relations one of these representations
must have to others, and the collection as a whole to the world, for it
to be a representation, say, of its raining. Then the teleological theorist,
call her Tilly, will come along and point out that surely some of the
causal roles of actual representations in actual people™s heads correspond
to bad inferences. What you must say, says Tilly, is that what the repre-
sentation represents is determined by what its causal role would be if the
head were operating correctly, that is, in the way it was designed, by
evolution or learning, to operate. Similarly, suppose that you think of
mental representations as items that “stand in for” the things they rep-
resent, running isomorphic to them, with differences in the representa-
tions producing differences in the behaviors guided by them, thus mak-
ing the behaviors appropriate to the presence of the things represented.
Then Tilly will come along and point out that some representations are
false, that is, not isomorphic to things in the world as required to guide
behavior appropriately. What you must say, says Tilly, is that the repre-
sentations represent what would be in the world, running isomorphic
to them, if the cognitive systems were operating correctly. That is, what
a teleological theory of content does is to take some more basic theory
of content, point out that the application of that theory to actual crea-
tures requires idealizing them in certain ways, and then offer the teleo-
logical principle to explain which idealization is the right one to use
in interpreting intentional contents, namely, the one that fits how the
cognitive systems were designed or selected for operating. You give

your naturalistic analysis of what a true or correct representation is like,
and Tilly merely adds that systems designed to produce true represen-
tations don™t always work as designed, claiming that correctness in per-
ception and cognition is defined by reference to design rather than ac-
tual disposition.
Accordingly, the teleologist who is an information semanticist begins
with the idea that representations are signals carrying “natural informa-
tion” and then adds teleology to account for error. My claim is that
adding teleology to informationL will not yield the rich variety of in-
tentional representation that either we or the animals employ, but that
there is a softer kind of natural information that does underlie all in-
tentional representation. This softer kind, however, offers no help what-
ever to the verificationist.
Let us return, for a few moments, to the animal whose perceptual/
cognitive systems are capable of translating informationL about the
same things arriving through a variety of media into a common code.
Whatever appears in that code is correlated always in the same way
with the same source or kind of source of informationL in the envi-
ronment. But, Tilly reminds us, it is not plausible that errors will never
occur. If this arrangement has been built by natural selection, however,
it will at least be a function of these mechanisms, which tap into and
converge these channels of informationL, to produce signals that carry
informationL in a univocal code. Their function is to transmit signals
that are controlled by certain external sources of information so that
these sources then control the behavior of the organism in ways that are
adaptive. Surely this is the sort of thing that Dretske had in mind in say-
ing that the function of a representation is to indicate (1986, 1991). Or,
being very careful, what has really been described here is not the func-
tion of the representations themselves, but the function of certain
mechanisms that produce representations. The first job of such a mech-
anism is to complete a specific type of channel of information flow, or
to bring to focus in a single code a number of such channels, so as to
produce an informationL-bearing signal in a specific code. This is the
way to add teleology to the idea that intentional representation is, at
root, natural informationL. False intentional representations result when
such a mechanism fails to perform this job properly.
I say that I think this is what Dretske has in mind. Dretske has some-
times wavered, however, on whether it can be a function of information
gathering systems to gather information about affairs that are distal to
the organism. I will explain.

The job of bringing information arriving through different channels,
perhaps through complex media, in different codes, to a focus is obvi-
ously difficult and very risky. Tilly is surely right that systems responsi-
ble for accomplishing this feat inevitably will sometimes fail. Recall the
Canada goose in love with itself, and the dog trying to communicate
with its echo. When this sort of thing happens, however, it is not usu-
ally because there is anything wrong with the organism.Without doubt,
perhaps definitionally, almost none of the mistakes in informationL
gathering that are made by healthy animals are due to malfunction of
the animals™ informationL-focusing systems. Mistakes are due to an un-
cooperative environment, which fails to supply those informationL
channels that the animal has been designed or tuned to recognize and
employ. Gibson to one side, concerning some informationL that an an-
imal needs to gather, the environment may be rife with decoy channels,
nor is there anything the animal can do about that, perhaps, without
evolving completely different perceptual systems. Both Dretske (1986)
and Neander (1995) have concluded from this, however, that the infor-
mation-gathering systems of animals may not actually have the function
of gathering information about distal affairs at all. The argument is that
when representations of distal affairs are apparently mistaken, since typ-
ically this is not because the animal™s information systems are failing to
function properly, it must be that these systems do not have as their
function to gather this kind of information. Neander then seriously
claims that all representation must be only of proximal stimuli. The ef-
fect, of course, will be a very strong form of verificationism indeed.The
organism can only represent what it can verify conclusively, granted it™s
not sick or damaged.
But the idea that nothing can have a purpose or function that it re-
quires help from anything else to achieve is mistaken. Consider the can
opener on the wall in my kitchen. It is not now opening cans. It is not
now performing its function. It would need my help in order to do
that. Certainly it doesn™t follow that it is malfunctioning, or that open-
ing cans is not its function.7 In the case of information-gathering sys-
tems, exactly as with can openers or, say, with the famous walking

7 If, however, you do insist, as Neander does, that in the ordinary sense of “function” things
really can™t have distal functions, then I refer you to the definition of “proper function”
stipulated in Millikan (1984), in accordance with which most of the many proper func-
tions that most biological items have are distal, and I suggest that the notion of function
we need to use to gain insight here is “proper function” as there defined.

mechanisms in cockroaches, a cooperative environment plays a lead role
in helping them serve their functions. (Nor, of course, does it follow
that it is the environment™s function to help cockroaches walk or to
help us focus information.)
Let us now look more closely at the result of adding teleology to
natural informationL to produce intentional representation.The first job
of a system that uses informationL to produce representations is to
complete a specific type of natural-informationL channel so as to pro-
ject that informationL into some standard code. But systems of this kind
also have jobs beyond. The codes into which they translate informa-
tionL must be ones that the behavioral systems of the animal are able to
use. The problem, posed first during evolutionary development, then to
the developing individual animal, is to coordinate these two kinds of
systems. Suppose, however, that the representation-producing systems
and the behavioral systems fail to cooperate on some task. Suppose that
a signal carrying informationL about one state of affairs is used by the
behavioral systems in a way appropriate instead to some contrary state
of affairs. For example, the informationL that the height to be stepped
up is, say, eight inches, is coded in a representation that guides the legs
to step up only seven inches.Which has erred, the perceptual side of the
system or the motor side of the system? Is the representation wrong, or
is its use wrong? Has the message been written wrong, or has it been
read wrong? What does the intentional representation say, eight inches or
seven inches?
Notice that the signal, as carrying informationL, definitely says eight
inches. Compare the informationL carried by a miscalibrated gas gauge.
The miscalibrated gauge carries informationL telling the actual level of
the gas in the tank. If we interpret it wrongly, that does not make it
carry the informationL we wrongly take it to carry. What it itself natu-
rally means just is whatever it actually carries informationL about, even
though in a difficult or uninterpretable code. In the same way, the
coded informationL about the height of the step cannot be wrong.The
attributes right and wrong, true and false, don™t apply to the code consid-
ered as a natural sign.
Recall that a signal carries informationL, not as considered within
the reference class of all items in the world having the same physical
form, but only as a member of the class of signals linked to sources
through the same kind of information channel, that is, in accordance
with the same natural necessities implemented through the same medi-
ating conditions. As an intentional representation, however, the represen-

tation of the height of the step is a member of a different reference class
altogether. It is a member of the class of all representations like it in
form,8 produced by the same representation-producing systems, for use
by the same representation-using systems. In this class there may also be
representations identical to it but that carry natural informationL in a
different code, and representations that carry no natural informationL at
all. In which code, then, is its intentional content expressed?
Exactly here is the place to apply teleology, as I see it, to the analy-
sis. We suppose that the system that codes and uses the information
about the step is a system where the coding and using parts of the sys-
tem have coevolved, either phylogenetically and/or ontogenetically.
During evolution of the species and/or during learning or tuning, they
have been selected or adjusted for their capacities to cooperate with
one another. The operative features of both halves of the system have
been selected for and/or tuned as they have because these features and
settings have sometimes succeeded in guiding behavior appropriate to
the informationL encoded. If this is so, inevitably it is true that these
coordinations were achieved by settling on some single and quite defi-
nite code. Only if there was constancy or stability in the code employed
by the representation maker and user could coordinations have been
achieved systematically. It is this code then that the representation pro-
ducer was designed to write in, and it is this code that the representa-
tion user was designed to read. And it is this code that determines the
intentional content of the message about the height of the step. In any
particular case of error, whether it is the representation producers or the
representation users that have erred depends on whether or not the nat-
ural informationL appears in this code.
My proposal is now that we should generalize this result. Intentional
representations and their producers are defined, are made to be such, by
the fact that it is their job to supply messages that correspond to the
world by a given code. That is the essence. But notice that that formula-
tion makes no reference to informationL. If that is the essence of the
matter, then the mechanisms by which the producers manage to produce
messages that correspond by the given code drops out as irrelevant to
their nature as intentional representation producers. If there exist systems
with the function of supplying messages that correspond to the world by
a given code but that manage to achieve this result, when successful,

8 More accurately, the class of all representations that the systems designed to use it are de-
signed to identify as having the same content.

without tapping into any channels of natural informationL, they too will be
producers of intentional representations. They will be producers of in-
tentional representations that are not defined with reference to natural
informationL. I will now argue that such systems do exist, indeed, that
the bulk of our mental representations necessarily are of this type.
Rather than informationL, they tap into channels of softer natural infor-
mation. How should we define this “softer” form of natural information?
Dretske wishes to eliminate de facto perfect correlations that are
“lucky coincidences” or “freaky piece[s] of good fortune” as possible
supports for any notion of natural information. But does anything stand
in the middle between, on the one hand, statistical frequencies resulting
from lucky coincidence and, on the other, the necessity of natural law?
The answer Dretske gave to this question, though inadequate, I believe,
is still a very interesting one. He said, “[t]here must actually be some
condition, lawful or otherwise, that explains the persistence of the corre-
lation” [emphasis mine]. About this I remarked earlier that the fact that
a local statistic is based lawfully upon prior local statistics, hence that a
correlation is explainable, does not alter its nature as a mere statistical
frequency. If the frequency of black balls in the urn today is 1, and if
nothing disturbs the urn, then by natural necessity it follows that the
frequency of balls in the urn tomorrow is 1. That does not change the
probability of being black if a ball in the urn into a probability of some
kind other than mere statistical frequency. It does not help being-a-ball-
in-the-urn to carry the informationL being-black.
But it does do something else. It explains how, by sampling the urn
today and adjusting my expectations of color accordingly, this adjust-
ment in expectation can turn out to be adequate to my experience to-
morrow, not by accident but for good reason. Many statistical frequencies per-
sist over time in accordance with natural necessity, and many produce
correlate statistical frequencies among causally related things, in accor-
dance with natural necessity. If measles are producing spots like that in
this community today, then measles will probably be producing spots
like that in this community tomorrow. Measles, after all, are contagious.
And if a nose like that is correlated with the presence of Johnny today
it will probably be correlated with the presence of Johnny tomorrow.
Johnny™s nose, after all, tends to sustain both its shape and its attachment
to Johnny. There are no laws that concern individuals as such, but there
are many kinds of local correlations that do. Notice, however, that
whether the persistence of a correlation may be explained in this sort

of way does not depend on its being a perfect correlation. Conditional
probabilities of 1 have nothing to do with the matter.
This yields a way that an organism may come to possess systems that
produce representations that correspond to the world by a given code
often enough to have been selected for doing that job, but that do this job
without tapping into any natural informationL. Systems of this sort run
on bare statistical frequencies of association “ on correlations “ but on
correlations that persist not by accident but for good reason. Probably
these correlations typically obtain between properties of the not-too-
distant environment that do supply informationL to the organism, and
more distal properties, kinds, situations, individuals, and so forth, of in-
terest to the organism but that don™t supply it with informationL. The
intentional contents of representations of this sort are determined not
by any natural informationL that it is their function to carry, but merely
by the codes in which their producers were selected to write, so as to
cooperate with the systems designed to read them.
It follows that a representation producer, basing its activities on past
local statistical frequencies, may indeed be representing Xs, and yet be
unable perfectly to distinguish Xs from Ys. It may have no disposition
under any conditions infallibly to distinguish Xs from Ys. To perform prop-
erly, its representations of Xs “ its code tokens of a certain type “ must
correspond to Xs, but this does not entail that there exist any informa-
tion channels at all, actual or possible, through which it could infallibly
discriminate Xs from Ys. That having grown up with gray squirrels
around, I am thinking of gray squirrels has nothing to do with whether
I can discriminate gray squirrels from Australian tree possums, even if
someone introduces tree possums into my neighborhood. Similarly, the
determinacy of content of my representation of cows is not threatened
by the possibility of a new species arising that I couldn™t distinguish from
cows, or by the possibility of Martians arriving with herds of facsimile
cows. The alternative that I should sometimes actually be at the other
end of an informationL channel from cows is not even coherent.
Consider, in this light, Pietroski™s tale about the kimus and the snorfs
(1992). The snorfs are attracted by the red morning glow over their lo-
cal mountain so that they climb up it each day. Thus they conveniently
avoid their chief predators, the snorfs, who don™t take to mountain ter-
rain. Pietroski claims that since no current kimu would recognize a
snorf if it ran into it head on, it is implausible that the perception of red
means snorf-free terrain to the kimus. A mere correlation between the

direction of the red glow and the direction of the snorfs is not enough
to support intentional representation. Now first, we should note that
the injection of phenomenology here is perversely distracting. The
question is not whether a red qualia, should there exist such things,
could mean no snorfs this direction rather than red. Bats perceive shapes by
ear and, goodness knows, maybe squares sound to them like diminished
seventh chords do to us. Pietroski™s question should be whether any in-
ner representation that merely directs the kimu toward the sunlight
could represent for it the snorf-free direction. Nor should the idea be
that the kimu reads or interprets the inner representation as meaning
“the snorf-free direction” the way you or I would interpret a sign of
snorfs. To interpret a sign of snorfs, you or I must have a prior way of
thinking of snorfs, and that, by hypothesis, the kimus do not have. The
question, put fairly, is whether something caused by red light could con-
stitute an inner representation of the snorf-free direction for the kimus.
Also, we should be clear that the kimus™ sensitivity to and attraction by
the red light is not supposed to be accidental, but is a result of natural
selection operating in the usual way. Kimu ancestors that were not at-
tracted to red light were eaten by the snorfs.
Put this way, the situation is parallel to that of certain tortoises, who
are attracted to green things, because green correlates with edible veg-
etation. They will move on the desert toward any green seen on their
horizon. Nor do the nutritious properties of the vegetation produce the
green light.These properties are merely correlated with green light. Can
the green mean “chow over there” to the tortoise? Obviously not in so
many words. But your percept of an apple doesn™t mean “there™s an
apple over there” in so many words either. If the green doesn™t mean
chow over there to the tortoise, then what on earth could mean chow
over there to anyone? Is it really plausible that there could be a genuine
informationL channel open to any of us, for you or for me, that would
communicate the informationL that there was chow on the table? Does
human chow, as such, figure in any causal laws? If not, then in what
sense are we “able to discriminate” when it is chow time? Unless, that
is, we rely on mere statistical correlations.
Besides natural informationL, then, we should recognize another
equally important kind of support for intentional representation, resting
on what may also be called “natural signs” carrying “ to keep the ter-
minology parallel “ “informationC” (for “correlation”). Natural signs
bearing informationC are, as such, instances of types that are correlated
with what they sign, there being a reason, grounded in natural necessity,

why this correlation extends through a period of time or from one part
of a locale to another. One thing carries information about another if it
is possible to learn from the one something about the other not as a
matter of accident but for a good reason. But no vehicle of information
is transparent, of course. How to read the information through its vehi-
cle has to be discovered, and it has to be possible to learn this in an ex-
plainable way, a way that works for a reason.The vehicle carries genuine
information only if there is an ontological ground supporting induction
that leads from prior experience to a grasp of the information carried
in new instances. There must be a connection between the various in-
stances exhibiting the correlation, a reason for the continuation of the
correlation. Correlations that yield true belief only by accident do not
carry genuine information.
Natural signs carrying informationC are correlated with what they
represent because each sign instance is connected with what it repre-
sents in a way that recurs for a reason. Typically, however, the correla-
tions are not perfect, and informationC, like informationL, cannot be
false by definition. A token indistinguishable from a natural sign but that
is not connected in the usual way with its usual represented is not a
natural sign. The correlations that support informationC may be weak
or strong. For example, a particular instance of a small shadow moving
across the ground is a natural sign carrying informationC that a flying
predator is overhead if it is actually caused by a flying predator, but the
correlation that supports this natural signing, though it persists for good
reason, may not be particularly strong.
If we allow ourselves to use the term “natural information” to cover
informationC as well as informationL, then, we must keep firmly in
mind that this sort of natural information has nothing to do with prob-
abilities of one. Nor does the presence of this kind of information di-
rectly require the truth of any counterfactuals. If a shadow is a natural

<< . .

( 29)

. . >>

Copyright Design by: Sunlight webdesign