Connectionism, Analogicity and Mental Content
The computational operations of an analog computer are driven by the natural causal
laws that apply to (aspects of)5 its material substrate. This substrate must possess sufficiently
complexity and variability to be capable of representing all of the relevant properties of the
target domain over which one wants to compute, together with their variation. This demands an
intimate relationship between the computer’s substrate and the representational vehicles it
implements, an intimacy that in practice results in a more continuous representational medium
(such that fine-grained variations in the substrate are representationally and hence
computationally significant). All this makes analog computation very different from digital
computation. In the latter case, a material substrate undergoes semantically coherent behaviour,
not in virtue of a structural isomorphism between its representational vehicles and its
representational domain, but by being forced to conform to a set of computational rules. Because
these computational rules are quite distinct from the causal laws that govern the behaviour of
the (continuously variable) physical properties comprising the material substrate, their
enforcement is achieved, as we have seen, through a syntactic partitioning of this substrate. It is
the satisfaction of this requirement that gives rise, in digital computers, to a more discrete
representational medium.
Analog computation is the dynamical alternative to Turing’s way. An analog conception
of cognition would thus represent an alternative to classicism. But what would such an
alternative look like? One answer, or at least I shall now argue, is connectionism.6
Whereas classicism is grounded in the computational theory underpinning the operation
of digital computers, connectionism relies on a neurally inspired computational framework
commonly known as parallel distributed processing (or just PDP).7 A PDP network consists in a
collection of processing units, each of which has a variable activation level. These units are
physically linked by connections, which enable the activation level of one unit to contribute to
the input and subsequent activation of other units. These connections incorporate modifiable
connection weights, which modulate the effect of one unit on another in either an excitatory or
inhibitory fashion. A PDP network typically performs computational operations by “relaxing”
into a stable pattern of activation in response to a stable array of inputs. Human cognitive
processes, according to connectionism, are the computational operations of a multitude of PDP
networks implemented in the neural hardware in our heads. And the human mind is viewed as
a coalition of interconnected, special-purpose, PDP devices whose combined activity is
responsible for the rich diversity of our thought and behaviour. This is the connectionist
computational theory of mind.
A strong hint that connectionism represents an analog conception of mind can be found
in the intimate relation between the PDP computational framework and the neuroanatomy of
the brain (something that is utterly lacking in digital models of cognition). Rumelhart and
McClelland, for example, have been explicit about the fact that PDP systems directly model
5 Not all of the physical properties of an analog computer's material substrate are relevant insofar as its computational
operations are concerned. It doesn't matter, for example, from what material a scale model of a proposed building is
constructed, as long as the material is opaque. This is because the physical property that implements the relevant
representational vehicle here is something like opaque shape (for want of a better term), and many different materials,
possessing all manner of different physical properties, can exhibit this physical property.
6 Connectionism represents one analog conception of mind, not the analog conception of mind (in contrast with the
way that classicism is the digital conception of mind). This is because connectionism is the analog theory of mind one
gets when one assumes that the neural network level of description of the brain is the right level for understanding
cognition. Other analog conceptions of mind, focusing on different levels of description, are conceivable (though
highly implausible, in my view).
7 The locus classicus of PDP is the two volume set by Rumelhart, McClelland, and the PDP Research Group
(Rumelhart & McClelland, 1986; McClelland and Rumelhart, 1986).