Varieties of perception 35
These two implications of Dretske’s view of information stand in an interesting but not immediately obvious relation to each other. Besides further inquiring
into the nature of natural information in the context of perception and aligning
the Dretskean with the Gibsonian concept of information, the meta-goal of what
follows in this chapter is to show how the two implications of Dretske’s view
are connected – although probably not in terms of being mutually supporting
constituents of his theory. It is the computationalist view of perception that is
ultimately committed (or condemned?) to relying on an objectivist standpoint in
which relations between a perceptual state and its object can be unambiguously
determined – while they cannot be warranted by the computational models proper.
Conversely, everything in the view from natural information, if and when pursued consistently, speaks for informational relations that are objective in kind but
context-bound all the way down, with no epistemic privilege to be assumed for
the human being perceiving and acting within an environment. In this chapter,
I will discuss three theories of perception that are markedly at variance with each
other in these respects: David Marr’s computational view of visual perception,
Gibson’s ecological approach to visual perception as the former’s main antagonist and the main focus of attention in this chapter, and the Empirical Strategy
of perception as a contemporary theory of intermediate status, which endorses
the notion of context-boundedness of perception while relying on computational
models. The issues of illusion and misperception discussed in that latter section
will be particularly instructive to elucidating the role of natural information and
its environmental context in perception.
Perception as information processing: the computational view
One peculiarity of all accounts of natural information discussed so far is that
they do not include any notion of data or information processing. Dretske, Brian
Skyrms, Ruth Millikan, like Gibson, all use concepts of natural information that
either rest on the expectation that all kinds of information are deducible from
a basic theory of natural information (Dretske, Skyrms) or do not include any
notion of likeness between natural information as used in perception and the
information-processing kind of information (Gibson, Millikan) – as Anthony
Chemero (2003b) has usefully highlighted.
Notions of data and information processing are characteristic of the nonsemantic mathematical theory of communication. They are also used in other,
non-naturalistic, semantic accounts of information, such as in “General Definition
of Information” or GDI that Luciano Floridi (2011) refers to as an “operational
standard” in fields such as data mining or information management. Floridi’s GDI
characterises semantic information as well-formed and meaningful data – a definition prima facie both more vague and more specific in character than Dretske’s.
It is more vague in having to be supplemented with further definitions capable
to cash out the requirements of being well-formed and meaningful. It is more
specific in focusing, by definition, on the kind of data that can be processed by
computers. Although not formally restricted to data of this kind, a fairly material
These two implications of Dretske’s view of information stand in an interesting but not immediately obvious relation to each other. Besides further inquiring
into the nature of natural information in the context of perception and aligning
the Dretskean with the Gibsonian concept of information, the meta-goal of what
follows in this chapter is to show how the two implications of Dretske’s view
are connected – although probably not in terms of being mutually supporting
constituents of his theory. It is the computationalist view of perception that is
ultimately committed (or condemned?) to relying on an objectivist standpoint in
which relations between a perceptual state and its object can be unambiguously
determined – while they cannot be warranted by the computational models proper.
Conversely, everything in the view from natural information, if and when pursued consistently, speaks for informational relations that are objective in kind but
context-bound all the way down, with no epistemic privilege to be assumed for
the human being perceiving and acting within an environment. In this chapter,
I will discuss three theories of perception that are markedly at variance with each
other in these respects: David Marr’s computational view of visual perception,
Gibson’s ecological approach to visual perception as the former’s main antagonist and the main focus of attention in this chapter, and the Empirical Strategy
of perception as a contemporary theory of intermediate status, which endorses
the notion of context-boundedness of perception while relying on computational
models. The issues of illusion and misperception discussed in that latter section
will be particularly instructive to elucidating the role of natural information and
its environmental context in perception.
Perception as information processing: the computational view
One peculiarity of all accounts of natural information discussed so far is that
they do not include any notion of data or information processing. Dretske, Brian
Skyrms, Ruth Millikan, like Gibson, all use concepts of natural information that
either rest on the expectation that all kinds of information are deducible from
a basic theory of natural information (Dretske, Skyrms) or do not include any
notion of likeness between natural information as used in perception and the
information-processing kind of information (Gibson, Millikan) – as Anthony
Chemero (2003b) has usefully highlighted.
Notions of data and information processing are characteristic of the nonsemantic mathematical theory of communication. They are also used in other,
non-naturalistic, semantic accounts of information, such as in “General Definition
of Information” or GDI that Luciano Floridi (2011) refers to as an “operational
standard” in fields such as data mining or information management. Floridi’s GDI
characterises semantic information as well-formed and meaningful data – a definition prima facie both more vague and more specific in character than Dretske’s.
It is more vague in having to be supplemented with further definitions capable
to cash out the requirements of being well-formed and meaningful. It is more
specific in focusing, by definition, on the kind of data that can be processed by
computers. Although not formally restricted to data of this kind, a fairly material