committed to anything about the inner organisation of the box, or could it
just as well be a distributed connectionist network? Consider what we
think takes place when someone engages in a simple piece of practical
reasoning: they want to eat an apple, see an apple in the fruit-bowl, and so
form and execute the intention to pick it up to eat it. Using the ab-
breviationsbel,desandintto represent belief, desire, and intention
respectively, and the square-bracket notation to represent contents, it
would seem that the folk are committed to the occurrence of at least the
following sequence of states:
des[I eat an apple]
bel[thereis an apple]
]int[I eat what isthere]There isWrst of all a commitment, here, to theco-occurrence of distinct
states of belief and desire which interact with one another to produce the
intention. But there seems no particular diYculty for connectionism in
this. It could be modelled by having two distinct banks of input nodes for
the network – one for belief-inputs and one for desire-inputs.
Second, there is a commitment to common conceptual components
between the states – it is because the desire is forapplesand my perceptual
belief represents anappleto betherethat I form the intention to eatthat
object (namely, the object which isthere). It seems unlikely that a dis-
tributed-connectionist system could preserve these properties. It seems
unlikely, in particular, that any given pattern of activation amongst the
input nodes will be preserved in the output nodes, such as would be
required for there to be a common conceptual component in thebeland in
theintstate, representing the location of the apple. Or rather, if there is
such a pattern, it will be entirely accidental, whereas folk psychology
considers this to be integral to, and necessary for the rationality of, the
inference. For if it were not thesamerepresentationWguring in thebeland
in theintstates, then we think that it would not be rational for me to act.
Admittedly, what folk-psychological explanation is committed to, in
the way of identity of representation, is sameness of conceptual content,
notnecessarily sameness of representational vehicle (as in a particular
sentence of Mentalese). Our argument is that robust and dependable
ways of processing identical conceptual contents are required. This eVec-
tively presents the advocate of connectionist cognitive architectures with
a dilemma. If a system of connectionist networkscannotdeliver these
shared conceptual contents in a dependable way, then it cannot model
the functional organisation of cognition – at least in so far as this is
known to folk psychology. On the other hand, if a system of such net-
workscandeliver such conceptual contents, then it becomes unclear why
Mentalese versus connectionism 201