<< . .

( 29)

. . >>

1. What reference classes are intended, then, when it is said that a cer-
tain r carries the information that a certain s is F? When Dretske says
that the pointer on my gas gauge being at the ½ mark carries the in-
formation that my gas tank is half full, in which two reference classes
are these two conditions being considered, so as to make that so?
Clearly the reference classes cannot be (1) all pointers on gas gauges
that point to the one half mark and (2) all gas tanks that are in the same
cars as those gauges. For some gas gauges are broken or disconnected or
badly calibrated, and even if none were, it would not be a matter of nat-
ural law that they couldn™t be broken or disconnected or badly cali-
brated. Rather, as Dretske emphasizes in KFI, a reference must be made
here to the presence of certain “channel conditions.” In this case, chan-
nel conditions consist in a fairly intricate collection of surrounding
conditions including various connecting parts the presence of which is
needed before natural laws will guarantee that the gas gauge will read
half full only if the gas tank is half full. One kind of thing carries in-
formation about another in accordance with strict natural necessity
only given specified channel conditions. The two reference classes con-
cerned contain only members connected by these channel conditions.
We can contrast this to the notion of a ceteris paribus law. According
to the classical view, a ceteris paribus law is one that is true in accordance

2 The necessity may go in either temporal direction. For example, an effect might carry in-
formation about another effect of the same cause.

with natural necessity given certain surrounding conditions, where ex-
actly what these conditions are is not specified, indeed, may or may not
be known. Usually the idea is, however, that whatever these conditions,
they are for the most part realized in the contexts in which the law is
used.The ceteris paribus law, then, makes reference to both kinds of prob-
ability that Dretske mentioned above. First, given the surrounding con-
ditions to which it implicitly refers, it holds true with a probability of 1
in accordance with strict natural necessity. Second, the surrounding
conditions to which it implicitly refers are themselves assumed to hold
true with high statistical frequency.
But on the above reading of Dretske™s definition of information, the
second sort of probability is not involved. The frequency with which
the channel conditions hold, relative to which a certain kind of signal
bears information about a certain kind of fact, is not part of the defi-
nition of information. Suppose, for example, that many gas gauges are
badly calibrated (indeed, they are) so that the gas tanks connected to
them are half full when the pointer is on the one quarter mark, others
when the pointer is on the three quarters mark, and so forth. In each
case, when the gas tank is half full, no matter what it reads, the pointer
carries the information that it is half full, relative to its own particular
channel conditions. How often each of these various kinds of channel
conditions holds is quite irrelevant. To be sure, Dretske often talks as if
the relevant reference class in which this reading on this gas gauge
should be put is restricted to those times when this very same gas
gauge does or, counterfactually, would have given this same reading.
Still, the assumption has to be that we are talking only about times
when this very same gas gauge is found surrounded by the very same
relevant channel conditions. Or suppose the reference class consists
only of this particular reading on this particular occasion, the idea be-
ing just that if the tank had not been half full the pointer would not
have pointed to this number. This way of thinking of the matter is in
every way equivalent. The point is that the counterfactuals have to be
run on the assumption that the relevant channel conditions still hold,
and nothing has been said about how often conditions of this sort do
hold in the world.
This is the only way I can see to interpret Dretske™s definition and
remarks on information quoted above. On the other hand, this way of
interpreting Dretske™s definition of information does seem to be incon-
sistent with certain things he says about “natural meaning,” “natural
signs,” and “indication” in EB, despite the fact that he explicitly associ-

ates all three of these with signals that bear “information” in the sense
of KFI (EB, p. 58). Dretske tells us, for example, that although otherwise
such tracks would indicate quale,“[i]f pheasants, also in the woods, leave
the very same kind of tracks, then the tracks, though made by a quail,
do not indicate that it was a quale that made them” (p. 56). Here, not
natural law but statistical frequencies at the source end of the informa-
tion channel appear to be determining whether the tracks carry natural
information. And Dretske tells us that “[t]he red spots all over Tommy™s
face mean [natural meaning] that he has the measles, not simply because
he has the measles, but because people without the measles don™t have
spots of that kind” (p. 56). Contrast Fodor, who seems to use the term
“information” more in the way we interpreted it above following the
explicit definition in KFI. He says, “If the tokens of a symbol have two
kinds of etiologies, it follows that there are two kinds of information
that tokens of that symbol carry. (If some ˜cow™ tokens are caused by
cows and some ˜cow™ tokens aren™t, then it follows that some ˜cow™ to-
kens carry information about cows and some ˜cow™ tokens don™t)”
(1990, p. 90). Fodor also often speaks of “covariation” between repre-
sented and representation, which is plausible only if one imagines a ref-
erence to some one definite though unspecified channel of influence,
making the signal depend nomically on whether s is F and vice versa.
Fodor™s usage fits not only Dretske™s original definition but also a cau-
tious physician™s offering: “Those spots may mean Tommy has the
measles, but they could also mean scarlet fever. I think we had better
take a culture.” Dretske™s modified claim, that if some people with spots
like that don™t have the measles then those spots don™t mean measles,
apparently refers instead to statistical frequencies at the source.
Alternatively, perhaps it refers to the frequency of certain channel
conditions. It might well be, for example, that given certain channel
conditions, only measles virus would cause spots like that, but that given
other channel conditions, only strep bacteria would. Just as, given cer-
tain channel conditions, only a half full tank of gas would cause that
reading, but given other channel conditions, only a quarter full tank
would.Then by Dretske™s original definition,Tommy™s spots might mean
measles even if on another child they would mean scarlet fever. But if
Dretske™s modification here involves assigning certain channel condi-
tions themselves a probability of one, such a probability would also
seem to be merely a statistical frequency.
Indeed, both Dretske™s KFI and his EB waver at points between the
two kinds of probability in discussing information. Dretske tells us, both

in KFI and in EB, that if his doorbell rings, that carries the information
that someone is at the door. But in EB we are told:

“It is partly the fact, presumably not itself a physical law, that animals do not
regularly depress doorbells . . . that makes a ringing doorbell mean that some
person is at the door. . . . As things now stand, we can say that the bell would not
be ringing if someone were not at the door. It therefore indicates or means that
someone is at the door. But this subjunctively expressed dependency, though
not a coincidence, is not grounded in natural law either. . . . Normally, though,
these things don™t happen. . . . And this is no lucky coincidence, no freaky
piece of good fortune. . . . There must actually be some condition, lawful or
otherwise, that explains the persistence of the correlation . . . [for the doorbell
to indicate a person].

But, of course, if the condition that explains the correlation is not law-
ful but “otherwise,” then it is grounded in mere facts about the state
conditions characterizing the world at certain times and places “ either
conditions at the source or existent channel conditions. It has the status
merely of a local statistical frequency “ based lawfully, perhaps, hence
explainably, upon prior local statistical frequencies, but that does not
change its essential nature as merely a statistical frequency.
The vacillation here seems to be twofold. First, it concerns whether
or not mere statistical frequencies at the source, rather than strict natural
law, should be allowed to determine signals as bearing “natural infor-
mation.” Second, it concerns whether we should count a signal that is
not univocal except as harnessed to a particular information channel.
But, of course, most of the interesting examples of signals carrying “in-
formation,” defined Dretske™s original way, are of a sort that either do
not always carry the same kind of information (because channel condi-
tions vary) or if they do, that is a matter of convenient empirical fact,
not natural necessity.The fact that a signal carries “information,” defined
Dretske™s original way, has no bearing whatever upon whether, by the
mere fact that the signal has arrived, one can tell anything about what
information, if any, it carries.3
I propose to stay for a while with Dretske™s original definition of nat-
ural information. To my knowledge, no other well-defined notion of
natural information is currently available. Allowing merely statistical

3 Dretske worries about something close to this in KFI, pp. 111“23, but he does so in
the confusing context of worrying about what “knowledge” is, and thus he never con-
fronts the basic problem. “ Or so I would argue, but my main project here is not Dretske

considerations on board poses an intractable problem concerning the
reference classes within which the frequency of 1 should be required to
hold. Do spots like that mean measles if small pox, though now extinct,
used to, and may in the future, cause spots like that? If the Skinner-
trained pigeons in your neighborhood start pressing doorbells, how
close may my neighborhood be to yours for my ringing doorbell still to
carry the information that a person is at my door? More important,
mixing frequencies with natural necessities muddies the issues involved
in trying to understand phenomena connected with intentional repre-
sentation. These issues can be seen much more clearly if we separate is-
sues of natural law from issues that concern mere frequencies. For clar-
ity, I will call natural information purified of all mere frequencies,
natural information as originally defined by Dretske, “informationL”
(for “law”).
InformationL is an entirely objective commodity and it is ubiqui-
tous. Often its channels are complex, and such as seldom if ever to be
duplicated. Channels that are often duplicated tend to be fairly simple
channels, such as reflections in calm water. Channels carrying reflections
in choppy water, though not much more complex, are seldom repeated.
The more numerous and irregular the intervening media between
source and signal are, the less likely repetition becomes.
InformationL is everywhere, but the problem, of course, is to inter-
pret it. For no signal that makes up only part of the world can carry the
informationL that its own channel conditions hold. And that means that
it cannot carry the information that it carries informationL, nor what
code this information is in. This opens the question why an organism
could possibly care whether or not it ever encounters any of this ubiq-
uitous but uncommunicative informationL.What good will it do an an-
imal to have informationL?
The problem is twofold. First, a signal carrying informationL is, as it
were, in code. It is of no use to an organism unless the organism can
“read” the code. Second, the informationL that reaches an organism is
not all in the same code.
Consider first the easy problem, that of reading the code. Suppose
that the information all arrives in the same code.Then for a signal to be
of use to a creature “ to be “read” by it “ it would only be necessary
that the creature should be guided by the signal in a way that diverts it
from activities less likely to benefit it to ones more likely to benefit it,
this likelihood being contingent on the fact conveyed by the signal. For
example, if the fact conveyed is the relative location of water, given that

the creature is thirsty, all that is needed is that the signal should cause
the creature to turn toward the location indicated. The beneficial activ-
ity need not, of course, be overt. It might be an inner state change.The
basic idea here is well known, I believe, and has been given numerous
expressions, for example, by Dretske and myself.
The “same code” problem is the harder one, and is itself two-sided.
First, we have not yet offered a reason to suppose that informationL
about the same thing always or ever reaches an organism in the same
code or packaging. Second, we have offered no reason to suppose that
the same packaging always or ever carries the same informationL, in-
deed, any informationL. Why suppose, for any signal that the organism
receives, that all signals of that kind reaching the organism, carry the
same informationL. But for the organism to be able to use the infor-
mationL it receives, the same kind of informational content needs to af-
fect the organism in the same kind of way, and different kinds of infor-
mational content need to affect it in different ways. Information about
the same must, as it were, look the same to the organism, and informa-
tion about different things must look different to the organism. (This
may put us in mind of Fodor™s “formality constraint” [1980].)
A central tenant of contemporary ecological psychology of the sort
introduced by J. J. Gibson is that there is far more consistency in the
natural information received by an organism than was formerly sup-
posed.The claim is, first, that if you look for the right aspect of the sig-
nals that arrive by way of the ambient energy surrounding an organism,
you find that a surprising number of superficially or apparently different
channels of informationL can be described as really being the same
channel once you have located the right high order invariances in the
signals. And it is these invariances, these univocal codes,4 that the evolv-
ing animal has become sensitive to, so as to “pick up” the relevant in-
formation and use it.
Second, the Gibsonian claim is that the very same relevant channel
conditions are present under environmental conditions that the animal
frequently or nearly always finds itself in, or that it knows how to ma-
neuver itself into. In the animal™s normal environment, the relevant
channel conditions are always the same, or always possible for the ani-

4 Gibsonians protest that the natural information used by organisms is not in the form of a
“code.”Their point, however, is merely that it is constituted by changing energy structures
that do not require translation into some other medium in order to be used by the

mal actively to intercept, so that relevant features of the source lawfully
produce informationL about themselves in the same code. There are
“ecological laws” such that the signals covary with the relevant envi-
ronmental features.
Third, the Gibsonian claim is that informationL of this sort that is
relevant to the animal™s needs is much more complete than had previ-
ously been supposed. Information about exactly those environmental
conditions to which the animal needs to adjust is frequently presented
in an unequivocal way. “The stimulus is not impoverished.”
These three claims are not generally separated in the tradition of
Gibsonian psychology, but they are independent. Gibsonian “informa-
tion” is not only informationL, but also lawfully carries complete infor-
mation needed for guidance with respect to important aspects of the
environment, and is frequently present in the environment, coming in
always through the very same information channels, that is, exemplify-
ing the very same ceteris paribus laws, arriving in a single code. All the
animal has to do is to tap into these rich sources of information (for ex-
ample, by developing eyes with lenses) and funnel them directly into
guidance of appropriate behavior.
Mechanisms by which various perceptual constancies are achieved,
such as recognition of same color, same shape, same size, same voice, and
so forth, through a wide spectrum of mediating conditions, insofar as
these constancies are sometimes detected over wide ranges of input in
accordance with univocal principles, illustrate the use of Gibsonian in-
formation. Then it is a very complex signal indeed, one in which the
significant invariances are (from the physicist™s point of view) highly de-
rived, that yields informationL through a complicated but still univocal
channel in a single code. The job of tapping into such informationL
channels and using the information to guide useful action is, as a bio-
logical engineering feat, extremely challenging.Yet natural selection has
managed to solve many of these problems.
Surely there does exist in our world at least a certain amount of Gib-
sonian information, or at least very close to that, which serves as the
bedrock foundation making sentient life possible at all. This foundation
guides the most basic immediate responses to the environment of all
animals, and also supports all information-gathering activities and fac-
ulties that make use of less tractable, less user-friendly, forms of infor-
mation that are also naturally found in the environment. But
there is also that useful information in the environment that is not
fully Gibsonian.

InformationL becomes less Gibsonian, for example, as it becomes less
ubiquitous. Darkness and dense fog, for example, impede transmission of
normal visual information. InformationL becomes less Gibsonian as it
arrives in more alternative packagings, in alternative codes. For example,
we use a number of alternative visual cues for depth. More interesting
are cases in which the same signal form varies in the information it car-
ries. Consider light and sound when reflected off smooth surfaces. Like
a gas gauge that carries informationL but reads “¼” when it is half full,
reflections carry perfectly good informationL but informationL that
needs to be read differently than usual. A puddle in the woods is not a
hole in the ground with upside down trees hanging inside. Animals, af-
ter brief exposure, generally treat reflections simply as irrelevant noise in
the data, holes in the normal flow of information. But a kitten™s first ex-
perience with a mirror can be very amusing to watch, a dog will bark
at its own echo, sometimes for hours, and a Canada goose once spent a
whole afternoon doing a mating dance to his reflection in the basement
window of our building on the Connecticut campus. We humans, on
the other hand, are able to tap many such sources of informationL and
to read them correctly. We can comb our hair in the mirror, we under-
stand that Clinton is not inside the TV set nor our friends inside the
telephone. We build gadgets to collect thousands of different kinds of
informationL “ various indicators, meters, gauges, scopes, audios, videos,
and so forth “ and we learn to read them correctly.
When a variety of channels of informationL about the same are in-
termittently available to an organism, the animal must understand when
each is open, distinguishing informationL both from mere noise and
from informationL arriving in similar vehicles but differently coded.
Nor should we take for granted that an animal can integrate the sources
of informationL that it uses. There is a story circulating (though proba-
bly apocryphal5) that certain venomous snakes strike mice by sight, trace
the path of the dying mouse by smell, and find its head (so as to swal-
low it first) by feel, and that none of these jobs can be done using any
other sensory modality. The lesson is, anyway, logically sound. Informa-
tionL about the same that comes in a variety of codes requires “transla-
tion” if it is to be used in a versatile way.
Suppose then that informationL about the same things arriving
through a variety of media is translated by mechanisms in the organism

5 The original source seems to be the zoologist Sverre Solander, who gives no references
and, despite requests, has offered no data yet to my knowledge.

into a common code.6 Insofar as this result is achieved, whatever ap-
pears in that code is correlated always in the same way with the same
source or kind of source in the environment, even when the channels
that control this effect are variable. In this way, a great deal of informa-
tionL that is not fully Gibsonian as it originally reaches the organism
may be translated into the practical equivalent of Gibsonian informa-
tionL inside the organism. As I will now argue, however, relatively few
things that an animal needs to know can be communicated in this di-
rect way.
InformationL depends on a channel between the information source
and the signal producing a correspondence between the two in accor-
dance with natural necessity. But unfortunately, relatively few things that
an animal needs to know about can figure as sources for this kind of in-
formation. The mouse, for example, needs to know when there is a
hawk overhead, but there are no natural laws that apply to hawks over-
head and hawks only.The existence of hawks is not a matter of law, nor,
for any given channel, is the nonexistence of things other than hawks
that might cause the same effects as a hawk on the output of that chan-
nel a matter of natural necessity. Similarly, if there are channel condi-
tions under which cows cause mental “cow” tokens as a matter of nat-
ural law, surely there can be none under which mental “cow” tokens are
caused by cows. They might instead be caused by something that
looked like a cow, or sounded like a cow, or smelled like a cow, or all
three, but that wasn™t a cow. It is the properties of objects like hawks and
cows that enter into natural laws, not the hawks and cows themselves,
and it is never a matter of natural law that only hawks or cows have
these properties.
There is, of course, an old-fashioned way out of this difficulty. You
can argue that it is a matter of nominal definition that cows and only
cows have certain properties, and then argue that information concern-
ing the copresence in one and the same object of all these defining
properties could indeed be transmitted through an information channel.
Then there might be natural informationL about the presence of a cow.
As a preliminary, however, first notice that you can™t take this route for
information concerning individuals. Even quite primitive animals are
often able to recognize and keep track of various of their conspecifics
individually, to learn things about them, and so forth. But there are no

6 By “translated into a common code,” I mean only that sameness or overlapping in content
is marked.

laws that concern any individuals as such. No signal can carry the in-
formationL that it is Tommy who has the measles. Second, although a
classical position that some still occupy gives natural kinds such as gold
and water definitions in terms of necessary and sufficient characteristics,
it is no longer plausible that biological kinds, such as cow, can be de-
fined that way. A large proportion of the kinds that we name in every-
day speech are “historical kinds,” kinds that are not defined by their
possession of certain properties at all, but instead through “historical”
connections “ connections in the spatial/temporal/causal order “ that
their members have to one another (Millikan 1999, Chapter 2 above).
Exactly as with individuals, these kinds cannot be subjects of informa-
tionL. They fall under no laws, not even ceteris paribus laws, and they
support no counterfactuals.
Thus we are returned to the problem addressed earlier when Dretske
observed that it is not a matter of natural necessity that your ringing
doorbell “indicates” there is some person at the door. In what sense of
“natural information” then, exactly, does the doorbell carry natural in-
formation? Is there a way to define a softer notion of “natural informa-
tion” to do the work required here?
To answer this we must have firmly in mind what work it is that is

<< . .

( 29)

. . >>

Copyright Design by: Sunlight webdesign