one of these domains requires an implausibly high number of training
runs, or an implausible degree of structure imposed upon the sequencing
of its inputs, to that extent it will fail as a model of human cognition in the
domain in question. The moral is: in order to assess any connectionist
claim, youWrst need to know the details of the training regime.
2.3 Connectionism and folk psychologyTo what extent is connectionism consistent with folk psychology? Ob-
viously that depends upon the commitments of the latter, and on the
nature, andWneness of grain, of the former. As will be clear from our
discussions in chapters 1, 2 and 4, we believe that folk psychology con-
ceives of the mind as a structured system which can be subjected to various
levels of functional analysis. So perception (in a number of diVerent
modalities) feeds into, but is distinct from, long-term belief; which in turn
is called upon to produce activated beliefs, judgements, and memories;
actions are produced by intentions and guided by beliefs and perceptions;
and intentions are the product of reasoning processes involving both
activated beliefs and activated desires, which are distinct kinds of state
from one another; and so on.
One way in which connectionism might be inconsistent with folk psy-
chology, then, would be by failing to replicate, to any signiWcant degree,
the right functional architecture. For example, the proposal that the brain
is a single large distributed network, within the operations of which it is
impossible to distinguish between perceptions, judgements and goals,
would surely be inconsistent with folk psychology. But then no one
seriously makes any such claim. Since it is known that the brain is sub-
divided into a huge variety of separate, functionally characterisable, sub-
systems, no connectionist should propose anything inconsistent with this.
In which case there will, to this extent, be no inconsistency with folk
psychology either.
To put these points somewhat diVerently: suppose we represented the
functional commitments of folk psychology boxologically, in aXow-chart
of diVerent processing systems. Then anyone (connectionists included) can
make proposals inconsistent with folk psychology by proposing functional
architectures which diVer from the commitments of the folk. But everyone
will surely agree to the minimal claim that there is some (complex)
functional architecture there to be described. And it would be possible for
connectionists to conWne their claims to the internal processes of the
various boxes in the diagram of folk belief.
Consider any one box within folk psychology’s postulated functional
organisation – the box for ‘practical reasoning’, say. Is folk psychology
200 Forms of representation