Resurrecting Dretskean information 27
thereby confirming that it indeed is a signal of that very affair in a given context.
If we briefly return to Dretske’s discussion of channel conditions, however, we
will see that the requirement looks slightly different: if the channel conditions are
variable, the animal will have to use a different, independent channel that either
confirms or disconfirms the original information or that transmits information on
the conditions in the former channel. Meta-information of this kind is separate
information and is to be assessed separately.
In either case, however, the consumer may not have additional means of independently verifying whether conditions hold under which a series of r-tokens or a
single instantiation of r in his or her stream of perceptions transmits the information that s is F, nor does he or she necessarily have independent means of verifying that conditions in his or her habitat are normal. Little knowledge on these
matters can be presumed on the side of the frog in our example. On a higher plane,
human cognition faces the same problem. In any given perceptual situation, we
cannot independently verify whether our perceptions are reliably connected with
their distal objects – which Dretske does not consider problematic because such
uncertainty does not affect the conditional probabilities of the informational relations as such (Dretske 1981, 56).
As an alternative to Dretske’s view and the problem of the mark of reliability
it seems to incur, Millikan (2001) suggests a decoupling of those probabilities
which hold for informational correlations at the source and those which apply to
the channel conditions, and thus to the noise and equivocation that will inevitably
enter into the transmission of the signal and hence reduce the reliability of the
information to the consumer. In prima facie analogy to Dretske, the probabilities
involved hence are treated separately again, but in a fashion that differs in one
important respect from Dretske’s account – namely in what status accrues to probabilities p < 1 for the co-occurrence of r, s and F.
In order to accommodate for the possibility of probabilities to be lower than
one in informational relations, Millikan goes on to distinguish between two types
of natural information: the first, christened “information L ” (or “context-free information” in Millikan 2004, 35), is based, in Dretske’s spirit, on correlations at the
source that hold with a probability of 1 by virtue of natural necessities. However, Millikan continues to argue, there are not many correlations of this kind
that would be within a normal consumer’s reach. Instead, all organisms rely on a
‘softer’ kind of information most of the time, which she christened “information C ”
(“local information” in Millikan 2004, 35). Information C is based on local statistical frequencies that may well remain below a probability of 1. These statistical
frequencies reflect a history of co-occurrences between signalling and signalled
affairs and thus are determined on a different base from Dretske’s in terms of
underlying regularities. These regularities are not nomologically but historically
and ecologically grounded. The statistical frequencies Millikan has in mind are
hence also determined on a different sample of occurrences, namely local, spatiotemporally bounded ones, and hence subject to variation.
Actually, the statistical frequencies for the signal to actually occur in relation to the signalled affair might even be extremely low without undermining
thereby confirming that it indeed is a signal of that very affair in a given context.
If we briefly return to Dretske’s discussion of channel conditions, however, we
will see that the requirement looks slightly different: if the channel conditions are
variable, the animal will have to use a different, independent channel that either
confirms or disconfirms the original information or that transmits information on
the conditions in the former channel. Meta-information of this kind is separate
information and is to be assessed separately.
In either case, however, the consumer may not have additional means of independently verifying whether conditions hold under which a series of r-tokens or a
single instantiation of r in his or her stream of perceptions transmits the information that s is F, nor does he or she necessarily have independent means of verifying that conditions in his or her habitat are normal. Little knowledge on these
matters can be presumed on the side of the frog in our example. On a higher plane,
human cognition faces the same problem. In any given perceptual situation, we
cannot independently verify whether our perceptions are reliably connected with
their distal objects – which Dretske does not consider problematic because such
uncertainty does not affect the conditional probabilities of the informational relations as such (Dretske 1981, 56).
As an alternative to Dretske’s view and the problem of the mark of reliability
it seems to incur, Millikan (2001) suggests a decoupling of those probabilities
which hold for informational correlations at the source and those which apply to
the channel conditions, and thus to the noise and equivocation that will inevitably
enter into the transmission of the signal and hence reduce the reliability of the
information to the consumer. In prima facie analogy to Dretske, the probabilities
involved hence are treated separately again, but in a fashion that differs in one
important respect from Dretske’s account – namely in what status accrues to probabilities p < 1 for the co-occurrence of r, s and F.
In order to accommodate for the possibility of probabilities to be lower than
one in informational relations, Millikan goes on to distinguish between two types
of natural information: the first, christened “information L ” (or “context-free information” in Millikan 2004, 35), is based, in Dretske’s spirit, on correlations at the
source that hold with a probability of 1 by virtue of natural necessities. However, Millikan continues to argue, there are not many correlations of this kind
that would be within a normal consumer’s reach. Instead, all organisms rely on a
‘softer’ kind of information most of the time, which she christened “information C ”
(“local information” in Millikan 2004, 35). Information C is based on local statistical frequencies that may well remain below a probability of 1. These statistical
frequencies reflect a history of co-occurrences between signalling and signalled
affairs and thus are determined on a different base from Dretske’s in terms of
underlying regularities. These regularities are not nomologically but historically
and ecologically grounded. The statistical frequencies Millikan has in mind are
hence also determined on a different sample of occurrences, namely local, spatiotemporally bounded ones, and hence subject to variation.
Actually, the statistical frequencies for the signal to actually occur in relation to the signalled affair might even be extremely low without undermining