to be considered neurologically plausible. For a model
to be neurologically plausible, it would need to deduce
new information about itself. More importantly, it would
be necessary to signal the newly obtained knowledge to
other neurons by changing the state of some nodes. Both
cluster analysis and current methods of KE clearly fail
to do this, although more recent developments in KE
can deliver much more accurate description of f. How-
ever the renewed and possibly accurate synthesis of re-
lations that were present in a training dataset does not
warrant claims that the ANN ‘discovered’, ‘learned’, or
‘recognized’ something or other, even if these relations
were not evident to the experimenter before. The abil-
ity to determine a function f that is contained in some
dataset illustrates the power of ANNs as analytical tools.
However, it should be clear that a different analytical
tool could also have been used to detect the function
f . We must conclude then that the model has failed to
explain any processes at the neural level. Instead, the
network model has only succeeded in offering an alterna-
tive method to encode the data, and the cluster analysis
provides an alternative method to analyze the data.
Conclusions
Computational models and simulations, and models us-
ing ANNs in particular, are commonly used in support of
theories about aspects of human cognition. Some mod-
els deal with high level psychological functions where
the operations at the neural level are of little interest,
and some models are concerned with the implementa-
tion of cognitive functions at neural level. I have ar-
gued that neurological possibility can be demonstrated
for nearly any conceivable psychological theory due to
the universality of simple ANNs. However using the lan-
guage and symbolism of neural nets does not support
any claims for neurological plausibility. The mistake,
I believe, is to bring the top-down psychological model
and the bottom-up neural environment together and to
treat the result as a coherent and meaningful demonstra-
tion. ANNs can be used successfully as models, provided
a clear description of the aims, assumptions and claims
are presented. However, when simple ANNs with small
numbers of nodes are employed to model complex high
level cognitive functions, the experimenter should eval-
uate whether the simplicity of the network can provide
a plausible implementation, because it is all too easy to
provide a neurologically possible model.
Acknowledgments
I would like to thank Anthony Corones for comments
and valued suggestions on earlier drafts of this paper.
References
Bennett, M. R. and Hacker, P. M. S. (2003). Philo-
sophical Foundations of Neuroscience. Malden, Mass.:
Blackwell.
proach of dealing with network structures is usually not em-
ployed by modelers either.
Branquinho, J. (2001). The foundations of cognitive sci-
ence. New York: Oxford UP.
Churchland, P. M. (1998). Toward a Cognitive Neurobi-
ology of the Moral Virtues. In Branquinho (2001).
Clark, A. (2001). Mindware: An Introduction to the
Philosophy of Cognitive Science. New York: Oxford
UP.
Cummins, R. and Delarosa Cummins, D. (2000). Minds,
Brains, and Computers: The Foundations of Cognitive
Science. Malden, Mass.: Blackwell.
Dretske, F. (1981). Knowlegde & the Flow of Informa-
tion. Cambridge, Mass.: MIT Press.
Dretske, F. (1988). Representational Systems. In
O’Connor and Robb (2003).
Elman, J. (1993). Learning and development in neural
networks: The importance of starting small. Cogni-
tion, 48:71-99.
Elman, J. L. (1990). Finding structure in time. Cognitive
Science, 14:179-211.
Elman, J. L., Bates, E. A., Karmiloff-Smith, A., Parisi,
D., and Plunkett, K. (1998). Rethinking Innateness:
A Connectionist Perspective on Development. Cam-
bridge, Mass.: MIT Press.
Fodor, J. and Pylyshyn, Z. (1988). Connectionism and
cogintive architecture: A critical analysis. Cognition,
28:3-71.
Green, C. D. (2001). Scientific models, connectionist
networks, and cognitive science. Theory & Psychology,
11(1):97-117.
Haugeland, J. (1981). Semantic Engines: An Introduc-
tion to Mind Design. In Cummins and Delarosa Cum-
mins (2000).
Haugeland, J. (1985). Artificial Intelligence: the Very
Idea. Cambridge, Mass.: MIT Press.
Hoffmann, A. (1998). Paradigms of Artificial Intelli-
gence. Singapore: Springer.
McLeod, P., Plunkett, K., and Rolls, E. T. (1998). Intro-
duction to Connectionist Modelling of Cognitive Pro-
cesses. Oxford: Oxford UP.
O’Connor, T. and Robb, D. (2003). Philosophy of Mind:
Contemporary Readings. London: Routledge.
Rogers, T. T. and McClelland, J. L. (2004). Seman-
tic Cognition: A Parallel Distributed Processing Ap-
proach. Cambridge, Mass.: MIT Press.
Rumelhart, D. E. and McClelland, J. L. (1996). On
Learning the Past Tense of English Verbs. In Cummins
and Delarosa Cummins (2000).
Schultz, T. R. (2003). Computational Developmental
Psychology. Cambridge, Mass.: MIT Press.
Uttal, W. R. (2001). The New Phrenology: The Limits of
Localizing Processes in the Brain. Cambridge, Mass.:
MIT Press.
1189