One typical functionalist category mistake

This post is also available in: Portuguese (Brazil)

The gadget engineer and entrepreneur-turned-theoretical neuroscientist Jeff Hawkins, author of On Intelligence, has a very interesting three-year old presentation which remains as one of my personal favorites out of TED Talks:

Jeff’s talk is filled with brain fodder (pun intended) such as his cognitivist criticisms of naïve behavioristic definitions of intelligence and his lucid appraisal of the current state of neuroscience as still dwelving in a Kuhnian phase of puzzle-solving, awaiting for a conceptual revolution. Defining intelligence in terms of the predictive capacity of an agent is as good as it gets for a prima facie definition of intelligence, as Dan Dennett once put it.

To me, one of the most striking features of this talk is a slide containing a diagram (in archetypical functionalist boxology) of what Mr. Hawkins sees as the main wrong-headed assumption behind Strong AI research programs:

That diagram couldn’t have been a more effective and explicit instance of what the epistemologist-turned-philosopher of mind Fred Dretske claimed to be a category mistake rampantly perpetrated by functionalists. It was one of the main themes of the first chapters of his intriguing and ambitious little book Explaining Behavior: Reasons in a World of Causes, published 22 years ago.

Dretske presents an ingenious metaphysical distinction between products (outputs) and processes, which are terms often used interchangeably even in the technical literature of artificial intelligence and philosophy of mind.

Processes aren’t just concatenations of states or events; a process is a temporally extended entity through which a certain outcome is terminated – meaning that a particular process ended. This output requires the through and through realization of each of the parts of the process, often in the right order. Imagine an automobile factory; if during the construction of a car you skip the painting stage, your product will be unfinished. And it wouldn’t make any sense at all to initiate the painting stage when one does not have any vehicle to paint. Processes can be constituted by other processes.

If a given product has not been deployed or realized, the process hasn’t finished. Processes do not cause products; processes are the production of products. And processes are partly constituted by their ending products.

Take for instance the process of delousing. The terminal condition of this process is the removal of some rather unpleasant hematophagous insects. However, strictly speaking, delousing does not cause dead lice. The bodily remains of the lice are precisely the product, which in turn are part of the process – the very last part of it.

For Dretske, behavior itself – which should be understood broadly, from breathing to thinking – is always a process token. Behaviors are processes endogenously caused in an organism or artifact which produce movements (again, this should also be understood broadly, not merely as visible movements but encompassing internal activities). And this is the functionalist category mistake – behavior is not the output of a system due to mereological considerations.

This applies even to Hawkins’ model and alternative framing of the issue; the most important outputs of the neocortex are predictions and expectations and these are themselves products of some sort of behavior, broadly considered. However, we can be charitable and interpret ‘behavior’ in his sense as external effects like speech acts and bodily motions.

And why should we care about this particular process/product distinction? Couldn’t this be just another useless and arid theoretical dispute? I’ll concede just one answer for now; I believe one is warranted to accept such a distinction in order to secure some autonomy of psychology from the biological sciences and the validity of psychological explanations at the personal level. Which is precisely the main philosophical project of Dretske in the rest of his book. I shall in the future discuss more of Dretske’s definition of processes, for instance its dependability on certain grammatical constraints.