112 Environments of intelligence
change, unless ending in failure of the entire process, will provide some degree
of variation that will result in a modified coupling relation further down the line.
Variability in environmental conditions will be one key factor in shaping coupling
relations.
The common rationale to unify this set of claims is an aetiological, that is historically based variety of functionalist argument. It departs from the classical Putnamian point of “machine-state functionalism”, which is based on the assumption
that one and the same function may be realised in a variety of different structures
(Putnam 1975a). This claim of multiple realisability has been central to any argument for the possibility of AI, and it is equally important to the “parity principle”
in the Extended Mind debates. It has its roots in Alan Turing’s theory of computability (see p. 3–4 in Chapter 1), which says that any logical or mathematical
operation that is formally solvable at all can be realised by elementary formal
operations that may be implemented in a broad, in principle indefinite, variety
of physical systems (1936). The claim of multiple realisability is also implied
by observations on neuronal equipotentiality, according to which one cognitive
or perceptual function can be realised by a variety of neuroanatomical structures
(see, e.g. the discussion of Karl Lashley’s work in Proust 1995).
While sharing with these types of functionalism the basic notion of a relation of
underdetermination between functions and structural properties of some system, an
aetiological account is interested in the concrete enabling and constraining conditions under which functions come to be established and realised. Functional aetiologies have their natural home in natural history and the Darwinian theory of evolution,
where evolved analogies of function between phylogenetically remote species have
been painstakingly distinguished from equally historically grounded, structural
homologies that do not necessarily have any bearing on functions at all (see p. 5 in
Chapter 1). Accordingly, the concrete histories of convergent and divergent functions
within and between populations will matter to an aetiological account.
If cognitive traits have biological functions in the same way as other traits of
an organism have biological functions, one will be entitled to analyse the mechanisms that realise these processes in the same functional-historical terms. Like
any other trait, they will be subject to processes of variation and natural selection.
In turn, the content of these processes can be analysed in analogous fashion, to
the extent that a type of cognitive state of one or a number of related individuals
is constituted by reproducible tokens that may succeed or fail to map onto some
world affair, and hence be selected to accomplish that mapping. These are the
basic assumptions of the teleosemantic or biofunctionalist paradigm in the philosophy of mind and language, as inaugurated by Ruth Millikan (1984).
4
More precisely, a functional-historical account of some cognitive trait will
recur to the contributions that the effects of that trait have made with sufficient
frequency among its ancestral bearers to their rate of reproduction as compared
with other members of the same population who did not possess that trait. By
virtue of conferring a reproductive advantage over the course of several generations, the trait will acquire the proper function of producing these effects (for the
full definition of proper functions, see Millikan (1984, 28 and Chapters 1–2 in
change, unless ending in failure of the entire process, will provide some degree
of variation that will result in a modified coupling relation further down the line.
Variability in environmental conditions will be one key factor in shaping coupling
relations.
The common rationale to unify this set of claims is an aetiological, that is historically based variety of functionalist argument. It departs from the classical Putnamian point of “machine-state functionalism”, which is based on the assumption
that one and the same function may be realised in a variety of different structures
(Putnam 1975a). This claim of multiple realisability has been central to any argument for the possibility of AI, and it is equally important to the “parity principle”
in the Extended Mind debates. It has its roots in Alan Turing’s theory of computability (see p. 3–4 in Chapter 1), which says that any logical or mathematical
operation that is formally solvable at all can be realised by elementary formal
operations that may be implemented in a broad, in principle indefinite, variety
of physical systems (1936). The claim of multiple realisability is also implied
by observations on neuronal equipotentiality, according to which one cognitive
or perceptual function can be realised by a variety of neuroanatomical structures
(see, e.g. the discussion of Karl Lashley’s work in Proust 1995).
While sharing with these types of functionalism the basic notion of a relation of
underdetermination between functions and structural properties of some system, an
aetiological account is interested in the concrete enabling and constraining conditions under which functions come to be established and realised. Functional aetiologies have their natural home in natural history and the Darwinian theory of evolution,
where evolved analogies of function between phylogenetically remote species have
been painstakingly distinguished from equally historically grounded, structural
homologies that do not necessarily have any bearing on functions at all (see p. 5 in
Chapter 1). Accordingly, the concrete histories of convergent and divergent functions
within and between populations will matter to an aetiological account.
If cognitive traits have biological functions in the same way as other traits of
an organism have biological functions, one will be entitled to analyse the mechanisms that realise these processes in the same functional-historical terms. Like
any other trait, they will be subject to processes of variation and natural selection.
In turn, the content of these processes can be analysed in analogous fashion, to
the extent that a type of cognitive state of one or a number of related individuals
is constituted by reproducible tokens that may succeed or fail to map onto some
world affair, and hence be selected to accomplish that mapping. These are the
basic assumptions of the teleosemantic or biofunctionalist paradigm in the philosophy of mind and language, as inaugurated by Ruth Millikan (1984).
4
More precisely, a functional-historical account of some cognitive trait will
recur to the contributions that the effects of that trait have made with sufficient
frequency among its ancestral bearers to their rate of reproduction as compared
with other members of the same population who did not possess that trait. By
virtue of conferring a reproductive advantage over the course of several generations, the trait will acquire the proper function of producing these effects (for the
full definition of proper functions, see Millikan (1984, 28 and Chapters 1–2 in