O'Brien (1998) and I (Green 1998) agree on many issues, but his reliance on Clark's (1990, 1993) justification for connectionist research in terms of explanatory inversion raises as many problems as it solves. Connectionism seems to generate ontological problems that do not impede symbolic cognitive science.
2. It first surfaces where O'Brien writes, "What [Green 1998] is highlighting is the fact... that connectionist explanations of cognitive phenomena are more dependent on implementational considerations than their more conventional ("classical") counterparts." I think I understand what O'Brien is getting at: whereas symbolic models do not seem to care whether they are implemented in brains or in beer cans and string (to use a particularly colorful example of Searle's, 1980), connectionists seem to take the fact that the only cognitive systems we know are implemented in brains far more seriously. The problem is, I am not sure I agree; or rather, I'm not sure I agree in the way O'Brien would want me to.
3. This "implementation argument" for connectionism (not to be confused with the implementationalism of Fodor & Pylyshyn 1988) seems to carry with it an implicit reductionist ontology that is difficult to credit in the 1990s. The "cognitive revolution" was all about undercutting reductionism. After 1970 or so, we began to see that one can be a "good" materialist about the mind without having to believe that the final descriptive vocabulary of psychology must be in terms of neurons, synapses, and the like. In other words, we saw how we could use beliefs, desires, and other terms of folk psychology as our fundamental vocabulary without having to reject materialism: such mental entities might "supervene" (Kim 1993) on their material substrate without being identified with particular material entities. Although it is true that I argue in the target article that connectionism is in a position such that it might have little or no content unless it identifies its units and connections with neural entities, that does nothing to undercut the supervenience argument in principle. Whether connectionism is the right theory of cognition is a question that must be fought on entirely separate grounds.
4. My concerns rose a little further when O'Brien brought in Andy Clark's claim that connectionism "inverts the usual temporal and methodological order of explanation" (Clark 1990, p. 299, cited by O'Brien 1998, para. 6). I have never understood what would motivate one to accept this claim, unless one had decided a priori, that connectionism must be the right theory of mind. Without that prior assumption, it seems to me that any argument seeking to radically change general Philosophy of science in order to solve a problem nested deeply within one particular science must, ceteris paribus, reduce the probability of that argument rather than increase it plausibility (for more on the same point, see my review of Clark 1993 in Green 1994). I also don't understand what is "Copernican" here. Copernicus is usually said to have radically simplified astronomical theory by placing the sun at the center (though this common claim is false: see Kuhn 1957). Clark's proposal, by contrast, seems to complicate matters, or at best leave them in more or less the same state they were with respect to complexity.
5. Naturally I would not go so far as to argue that we should prevent connectionists from experimenting with networks that do not satisfy the requirement of adhering doggedly to all we now (think we) know about neural function (O'Brien 1998, para. 8). By the same token, I think we should look upon such work with a healthy amount of scepticism. It is connectionism that put itself in the ontological box I outlined in the target article. If it succeeds in surviving there over the long term, it will be considered a great victory. In the meantime, however, symbolic cognitive science is under no obligation either to help connectionism out of the box if it is in trouble, nor to forswear its own more liberal ontology in order to "level the playing field."
Clark, A. (1990) Connectionism, competence, and explanation. In: The Philosophy of Artificial Intelligence, ed. Boden, M. Oxford: Oxford University Press.
Clark, A. (1993) Associative engines: Connectionism, concepts, and representational change. MIT Press.
Kim, J. (1993). Supervenience and mind_. Cambridge University Press.
Fodor, J. A. & Pylyshyn, Z. W. (1988) Connectionism and cognitive architecture: A critical analysis. Cognition 28: 3-71.
Green, C. D. (1994) Fighting the good fight: A review of Andy Clark's Associative engines: Connectionism, concepts, and representational change. Canadian Artificial Intelligence, No. 35, 33-36. http://www.yorku.ca/faculty/academic/christo/papers/clarkrev.htm
Green, C. D. (1998) 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.
Kuhn, T. S. (1957). The Copernican revolution: Planetary astronomy in the development of Western thought. Harvard University Press.
O'Brien, G. J. (1998) The role of implementation in connectionist explanation. PSYCOLOQUY 9(6). ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1998.volume.9/ psyc.98.9.06.connectionist-explanation.3.obrien.
Searle, J. R. (1980) Minds, brains and programs. Behavioral and Brain Sciences 3: 417-424.