1.1.3 Fodor and Pylyshyn on Connectionist Networks
Another famous Hegelian argument is the famous systematicity argument
found in ‘‘Connectionism and the Cognitive’’ Architecture by Fodor and
Pylyshyn
6
(1988; see also Aizawa 2003, Fodor 2008). Their argument
against connectionist networks as a model of the cognitive architecture
goes as follows.
1. Human thought is systematic. That is, abilities come in clusters.
2. Systematicity requires representations with compositional structure.
3. Connectionist networks do not have representations with composi-
tional structure.
4. Therefore, connectionist networks are not good models of human
thought.
This argument is one of the most important and influential in the recent
history of cognitive science. It drew stark battle lines soon after Rumelhart,
McClelland, and the PDP Research Group (1986) drew attention to connec-
tionist networks. It is also an argument that has been convincing to many
people. That this argument is Hegelian can be seen from premises (1) and
(3), neither of which is defended in Fodor and Pylyshyn’s article by citing
empirical studies. In fact, Fodor and Pylyshyn’s claim that human thought
is systematic was an entirely new one in the cognitive sciences. No prior
empirical study supported the claim that human thought came in ‘‘clus-
ters,’’ where having one ability was necessarily connected to having others.
In fact, Fodor and Pylyshyn cite just one empirical study in the whole of
their paper: a chapter in Pinker 1984. In that chapter, the evidence pre-
sented indicates that children’s speech is not systematic, which evidence
Pinker attempts to discount. So the only experimental study cited actually
contains evidence against systematicity.
7
And, in fact, Dennett (1991) and
Clark (1997), among others, have argued (against premise 1) that although
human language is systematic, the rest of human thought is not. Many
other defenders of connectionism, such as Smolensky (1990), van Gelder
(1990), and Chalmers (1990), have argued (against premise 3) that connec-
tionist networks can have representations with compositional structure.
My reminder of these two highly respected, highly influential Hegelian
arguments from cognitive science is intended to show that these arguments
are taken very seriously in the discipline. This is an important difference
between cognitive science and other sciences, which are more purely em-
pirical. At first blush, it might seem that the explanation for this differ-
ence is that the cognition is different in kind from the subject matter of
other sciences, and requires a different kind of science. Is this true? To see
8 Chapter 1

III Ecological Psychology
The rules that govern behavior are not like laws enforced by an authority or decisions
made by a commander: behavior is regular without being regulated. The question is
how this can be.
—James J. Gibson, The Ecological Approach to Visual Perception (1979)