Christopher D. Green (1998) Localist Connectionism Does not Address the Issue. Psycoloquy: 9(27) Connectionist Explanation (24)

Volume: 9 (next, prev) Issue: 27 (next, prev) Article: 24 (next prev first) Alternate versions: ASCII Summary

Topic:

Article:

PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).

Psycoloquy 9(27): Localist Connectionism Does not Address the Issue

LOCALIST CONNECTIONISM DOES NOT ADDRESS THE ISSUE
Reply to Watters on Connectionist-Explanation

Christopher D. Green
Department of Psychology
York University
Toronto, Ontario M3J 1P3
Canada
http://www.yorku.ca/faculty/academic/christo

christo@yorku.ca

Abstract

Watters (1998) is concerned that I excluded localist connectionist networks from consideration (Green 1998a). As indicated in my prior reply (Green 1998b), localist nets do not suffer from the problems outlined in the target article because they are a species of "classical" symbolic cognitive theory.

Keywords

artificial intelligence, cognition, computer modelling, connectionism, epistemology, explanation, methodology, neural nets, philosophy of science, theory.

1. Watters (1998) begins with a brief reminder about the history of connectionism (I hesitate to use the term "neural networks," because it begs the very question raised in Green, 1998a). He then suggests that in conjunction with (traditional symbolic?) theories of cognition, parallel distributed connectionist networks can be useful exploratory tools: "useful in the same way that polynomial regression can be useful in modelling nonlinearities in data whose underlying structure is unknown" (para. 4). Finally, he presents some examples in which LOCALIST connectionist networks have served as useful theories of cognitive processes.

2. Watters's final point is an interesting one, but it is not of much concern to my argument. It is not so much that localist nets were an alternative that I failed to consider (Watters 1998, para. 5). It is, rather, that localist nets are not germane to my argument. Although I may have failed to say so explicitly (expecting that it would be clear from my examples), my main concern was with parallel distributed connectionist networks, not localist ones. As argued in a reply to an earlier commentary (Green 1998b), localist networks are a species of "classical" symbolic model, and as such suffer from the difficulties outlined in the target article no more than any other symbolic model. Some may wish to question whether concepts, beliefs, and desires are an appropriate ontological basis on which to develop a cognitive theory, but that matter is entirely distinct from the one raised in the target article. Distributed connectionist networks do not have an ontological base even as strong as that.

3. As to Watters's second point, that parallel distributed networks can be seen as valuable TOOLS (rather than MODELS, per se) to explore traditional symbolic cognitive theories, I am in agreement. I did not say that connectionism was of no use at all, only that (distributed) connectionist networks of cognitive processes suffer from serious difficulties when regarded as theories of those processes.

REFERENCES

Green, C.D. (1998a) Are connectionist models theories of cognition? PSYCOLOQUY 9(4) ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1998.volume.9/ psyc.98.9.04.connectionist-explanation.1.green

Green, C.D. (1998b) Does localist connectionism solve the problem? Reply to Grainger & Jacobs on Connectionist-Explanation PSYCOLOQUY 9(14) ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1998.volume.9/ psyc.98.9.14.connectionist-explanation.11.green

Watters, P.A. (1998) Cognitive theory and neural model: The role of local representation: Commentary on Green on Connectionist-Explanation. PSYCOLOQUY 9(20) ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1998.volume.9/ psycoloquy.98.9.20.connectionist-explanation.17.green