Minds and Computers : An Introduction to the Philosophy of Artificial Intelligence

mental representations, which are then involved in further computa-
tional processes – such as the comprehension of natural language
utterances.
There is a distinction to be drawn here between the conditions for
the development of semantic representations and the conditions
under which tokens of these representations are held to be meaning-
ful. In other words, we might concede the necessity of embodied
experience for the development of semantic representations, but then
consider a thought experiment where functional equivalents of the
formal system(s) facilitating language production and comprehen-
sion in a fully developed native speaker are enacted in a way approx-
imating the Chinese room example, hoping to further problematise
the meaningfulness of the operations of such a system.
Recall, however, from Chapter 14 that although there is no reason
to suppose that these linguistic processes are not computational, there
is good evidence to support the claim that concerted appeals to
various kinds of mental representations – including semantic repre-
sentations – are a necessary feature of their operations.
In other words, it is not the case that the processes underwriting lin-
guistic comprehension are isolated, modular processes, with meaning
being assigned as the final stage of processing. Rather, these various
processes occur in concert with appeals to semantic representations
serving to constrain and inform phonological and syntactic processes.
Consequently – although I’ve o
ffered neither proof nor robust argu-
ment here – it seems likely that any empirically possible system
su
fficient for passing a Turing test will necessarily already contain
meaningful semantic representations.
There is much more to be said about the conditions under which oper-
ations of a formal system can be held to be intrinsically meaningful
(not just interpretable as meaningful). In the following chapter, I want
to shift the focus of our examination of meaningfulness to explicit
consideration of this notion of mental representation.
180
  

---

C H A P T E R 1 8
REPRESENTATION
Our mental states are meaningful by virtue of being about things.
In other words, meaningful mental states are representational states –
they represent or stand for things. In previous chapters I’ve made
reference to mental representations, such as the phonemic, syntactic
and – crucially – semantic representations which facilitate linguistic
production and comprehension. In this chapter, I want to briefly
discuss the structure and nature of mental representation.
Representation is quite a thorny philosophical topic. I don’t intend
this chapter to be a comprehensive introduction to the various debates
concerning mental representation. Rather, I want to use this discus-
sion of the nature of mental representation to introduce a distinction
between two competing paradigms in artificial intelligence research.
The most important distinction, for our purposes, between these
two paradigms – which I will call the symbolic and the connectionist
paradigms – lies in the methods deployed by researchers in trying to
computationally replicate cognitive functions. The symbolic artificial
intelligence researcher will employ symbol systems of the kind we are
now very familiar with, having seen numerous examples in previous
chapters. The connectionist artificial intelligence researcher, on the
other hand, will construct artificial neural networks.
We’re going to examine artificial neural networks – or connection-
ist networks – at length in Chapter 19. In what follows, I want to make
clear that this distinction in artificial intelligence methodology is