Did McCulloch and Pitts 1943 Offer a Theory of Cognition in terms of Digital Computation?
Posted Nov 26 2009 10:00pm
Some people don't realize this. Someone recently objected to me that since McCulloch and Pitts were studying the brain, they must be offering only a theory of neural activity, on the same level as, say, Hodgkin and Huxley. How can I say that McCulloch and Pitts had a theory of cognition?
Although McCulloch and Pitts did not use the term "cognition," which became widespread only later, they were explicitly attempting to explain mental phenomena in terms of (highly idealized) neural activity. In describing what they were explaining, they used terms like "mind," "knowledge," "ideas," and "purposes". So in modern terms, they had a theory of cognition.
Furthermore, the same person who objected to the theory of cognition part also objected that McCulloch and Pitts were connectionists, so they couldn't be explaining cognition in terms of digital computation.
But this objection is just based on confusion. In the broad sense of the term "connectionism" (meaning explanation in terms of networks of simple processing units), there is no contrast between being a connectionist and appealing to digital computation. Digital computers themselves are just giant connectionist systems! In a more narrow sense of the term "connectionism" (meaning explanation in terms of learned associations between neurons), McCulloch was explicitly anti-connectionist. McCulloch was quite hostile to Roseblatt's work. Rosemblatt, of course, is the guy who worked on associative networks called "perceptrons" and introduced the term "connectionism" in the modern literature on neural networks.
Does any of this matter? I think it does, insofar as the continuing misunderstanding of McCulloch and Pitts's landmark theory continues to be a drag on our understanding of the origin and nature of the computational theory of mind. For my thoughts on McCulloch and Pitt's theory and its significance, see this paper.
NB: Ken Aizawa has legitimate qualms about the extent to which digital computation played a central role in McCulloch and Pitts's actual theory at the time it was proposed. I am not taking issue with Aizawa's qualms. But I do stress that regardless of the exact role played by digital computation in McCulloch and Pitts's theory, their paper did play the historically very significant role of injecting digital computation into the theory of cognition (and neural activity).