58 Informational environments
will be an external observer by default, whereas in cases NI-2b and NI-2c, the
receiver will typically be involved in a concrete situation of tracking information
that is relevant to him. Conversely, what is relevant to him and what he is capable
of tracking will make a difference to the probabilities of the signals he uses to
actually match distal conditions. The issue of the level on which and the degree
to which the receivers or “consumers” of natural information become important
is hence related to the issue of the degree to which natural information may be
an openly probabilistic affair, that is an affair allowing for probabilities p < 1.
These two points of contention concern the proper choice of a framework of reference for determining the probabilities of informational relations (see p. 23 in
Chapter 2).
Already quite early in the discussion of Dretske’s (1981) information-theoretical account of knowledge, Gilbert Harman (1983) pointed out that one could
determine the probability of a certain outcome in any information-generating situation only in relation to a given “framework”, so as to measure the reduction
of possibilities that takes place – a point not contested by Dretske (1983, in
his “Author’s Response” to Harman’s peer commentary). The observation of a
coin-tossing game with a set of unequally fair coins, for the mixed probabilities involved, will deliver different values than the observation of frequencies of
results for each coin individually. Similarly, Dretske’s requirement of p = 1 for
informational content can be met only if the framework or “reference class”, as
Dretske himself calls it, is tailored tight enough to allow only this outcome, so
that, whatever the initial distribution of possible states has been, the result is unity.
The observation of a coin-tossing game, if comprising a reasonably large sample
and if confined to one or the set of all fair coins, will never produce a conditional
probability p > .5 over the course of that observation. Here, the reference class is
the entire game or set of games, as observed over a number of rounds between t 0
and t n . However, quite obviously, we will know with certainty what result only
one toss in the series produces after we have observed that one toss of the coin.
Here, the reference class will be restricted to one single case, where this case, in
turn, is supposedly governed by laws of nature – that set of laws which is sufficient to explain why the coin, under the given set of antecedent conditions, landed
this way.
Cases of this latter sort are what Dretske envisions. Only under these conditions, Dretske’s underlying argument goes, information may convey knowledge.
There would be no knowledge if conditions at the source were allowed to be at
variance with what is being signalled with whatever small degree of likeliness.
On these grounds, he restricts the assignment of the status of information to those
cases where the condition of unity is met. Probabilities concerning the entirety
of conditions within sets of events may be shaped by a variety of factors, and are
ultimately not his concern.
As authors such as Millikan and Skyrms argue with some justification, setting
admission criteria to the realm of information as strict as Dretske might help us to
an account of information that satisfies a (foundationalist) criterion of knowledge
but bears little information on how organisms successfully perceive and act under
will be an external observer by default, whereas in cases NI-2b and NI-2c, the
receiver will typically be involved in a concrete situation of tracking information
that is relevant to him. Conversely, what is relevant to him and what he is capable
of tracking will make a difference to the probabilities of the signals he uses to
actually match distal conditions. The issue of the level on which and the degree
to which the receivers or “consumers” of natural information become important
is hence related to the issue of the degree to which natural information may be
an openly probabilistic affair, that is an affair allowing for probabilities p < 1.
These two points of contention concern the proper choice of a framework of reference for determining the probabilities of informational relations (see p. 23 in
Chapter 2).
Already quite early in the discussion of Dretske’s (1981) information-theoretical account of knowledge, Gilbert Harman (1983) pointed out that one could
determine the probability of a certain outcome in any information-generating situation only in relation to a given “framework”, so as to measure the reduction
of possibilities that takes place – a point not contested by Dretske (1983, in
his “Author’s Response” to Harman’s peer commentary). The observation of a
coin-tossing game with a set of unequally fair coins, for the mixed probabilities involved, will deliver different values than the observation of frequencies of
results for each coin individually. Similarly, Dretske’s requirement of p = 1 for
informational content can be met only if the framework or “reference class”, as
Dretske himself calls it, is tailored tight enough to allow only this outcome, so
that, whatever the initial distribution of possible states has been, the result is unity.
The observation of a coin-tossing game, if comprising a reasonably large sample
and if confined to one or the set of all fair coins, will never produce a conditional
probability p > .5 over the course of that observation. Here, the reference class is
the entire game or set of games, as observed over a number of rounds between t 0
and t n . However, quite obviously, we will know with certainty what result only
one toss in the series produces after we have observed that one toss of the coin.
Here, the reference class will be restricted to one single case, where this case, in
turn, is supposedly governed by laws of nature – that set of laws which is sufficient to explain why the coin, under the given set of antecedent conditions, landed
this way.
Cases of this latter sort are what Dretske envisions. Only under these conditions, Dretske’s underlying argument goes, information may convey knowledge.
There would be no knowledge if conditions at the source were allowed to be at
variance with what is being signalled with whatever small degree of likeliness.
On these grounds, he restricts the assignment of the status of information to those
cases where the condition of unity is met. Probabilities concerning the entirety
of conditions within sets of events may be shaped by a variety of factors, and are
ultimately not his concern.
As authors such as Millikan and Skyrms argue with some justification, setting
admission criteria to the realm of information as strict as Dretske might help us to
an account of information that satisfies a (foundationalist) criterion of knowledge
but bears little information on how organisms successfully perceive and act under