Commentary from Dimitri Coelho Mollo on today’s post from Mazviita Chirimuuta on The Brain Abstracted (MIT Press).
I was lucky to have had the chance to discuss this brilliant book with Mazviita Chirimuuta and others while it was still in preparation, and I’m looking forward to exchanging ideas about it once more over here at the BrainsBlog! I will focus especially on her treatment of cognitive representations, which occupies chapter 6 of the book, and whose spirit Chirimuuta summarises in today’s post. (Actually, she talks about neural representations, a terminological choice I somewhat dislike, as it suggests a reductionist-mechanistic approach to representations in cognitive science that both of us reject).
Chirimuuta’s take on cognitive representations finds me largely in agreement (see my 2022). We both think that representational explanation in cognitive science draws on analogies to external, public representations. We both want to point out that, as with any analogy, the fit between the two domains being compared is partial, and differences abound. The attempt to fit cognitive representations too tightly into the analogy with public representations has led, I’ve argued, to misplaced metaphysical worries about things such as content determinacy, and the individuation of representational vehicles, consumers, and producers. Like Chirimuuta, I also think that representational explanations, as most if not all explanations in science, involve a fair amount of abstraction and idealisation.
For the rest of this post, I will focus though on three main points of disagreement (I’m a philosopher, after all!), which I find are quite interesting in the differences they reveal about how we see current work on cognitive representation; and how much weight we give to certain argumentative strategies in philosophy of science, such as the move from the explanatory success of a scientific posit to a more or less cautious realism about it.
First, I’m much more optimistic about the project of naturalising representation than Chirimuuta seems to be. She mentions that there is no consensus view as yet, which is true. But this masks the fact that there is wide agreement (at least as wide as these things get in philosophy) that something along the lines of informational teleosemantics is right. In other words, there is quite some consensus that content is a matter of causal relations and teleological functions: disagreements persist mostly in how to cash out the details (cf. Neander 2017, Millikan 2017, Shea 2018).
Moreover, although I have occasionally, and regrettably, suggested that the naturalisation project is in the business of ‘saving’ cognitive science by providing philosophical load-bearing columns to secure its foundations—something that Mazviita rightly criticises in the book—I don’t (now) think that that should be a guiding motivation nor justification for that project. Rather, it’s just an ordinary instance of philosophy of science: scientists use posits that prove to be fruitful and explanatory, such as cognitive representation (and computation); and philosophers of science then get to fuss about the nature of such posits, how to make them cohere with our current state of scientific and philosophical knowledge, and so on. Naturalising representation, I think, should be seen on a par with attempts to find the best way to understand posits such as genes, species, and the like. It may turn out that that is the wrong way to go, and something like fictionalism or pragmatism about representations is correct (I doubt it, though!). But the legitimacy of scientific appeal to representational explanation does not ride on the success or else of naturalisation.
Second, I don’t quite agree that the representational strategy in cognitive science is mainly a matter of simplification. And this for reasons that Chirimuuta mentions both in her post and in the book. The explanatory targets of cognitive neuroscience are relational in nature: we try to figure out what explains certain sorts of environment-sensitive behaviour on the part of (for now only biological) systems. This makes it unavoidable to focus on the relationships between (multiply realised) distal items and (multiply realisable) system states, and to ignore the implementational details of the (token) causal chains. To do otherwise would be to explain something else, rather than cognitive or intelligent behaviour (pace proponents of mechanistic explanatory monism). Cognitive representations are posited as a way to explain such relations, and thus to capture what is relevant to explaining cognitive behaviour. Simplification is then at best a fortunate side-effect, and not the key guiding force of representational explanations.
Third, and finally, Chirimuuta maintains metaphysical neutrality about the status of cognitive representations, focusing just on their epistemic benefits. I wonder though to what extent those things can go together, if, as Chirimuuta recognises, cognitive representations are extremely fruitful posits, appear in lots of successful explanations, and may even be fundamental to cognitive scientific explanations (cf p. 167 of the book). These considerations strike me as strongly indicative that we should prefer a cautious realism about cognitive representations. Sure, idealisations and abstractions are at play, so we don’t get a perfect model-reality fit (do we ever?), but we do get to capture central features of how cognition works, to a (very) good approximation. Maybe this is a way of being a haptic realist? At any rate, I think Chirimuuta gives very good reasons to be some sort of realist about cognitive representation, making her stated metaphysical neutrality a problem, rather than a strength of her view.
References
Coelho Mollo, D. (2022). Deflationary Realism: idealisation and representation in cognitive science. Mind & Language 37(5):1048-1066.
Millikan, R.G. (2017). Beyond Concepts: unicepts, language, and natural information. OUP.
Neander, K. (2017). A Mark of the Mental. MIT Press.
Shea, N. (2018). Representation in Cognitive Science. OUP.
Dimitri makes the case for a moderate realism about cognitive representations, as part of a broader project of naturalizing intentionality. Some of the points of disagreement can be put down to me focusing to a large degree on representations in sensory neuroscience, whereas Dimitri has neuroscience of cognition more globally in view. Also, I say something very similar to Dimitri regarding the second point: the explanatory targets of cognitive neuroscience are the relations between neural activity and distal events. In order to have those relations in focus you need to abstract away from more proximal causal influences on neural activations, and I argue that the analogy with public representations permits this abstraction. So, it is an abstraction performed for the very purpose that Dimitri highlights. Perhaps this is not the right moment for frank confession, but I feel I should say something more about my underlying motivations in this chapter. An overall theme of the book is that the mind/brain sciences have suffered under importation of a physics-based model of how to undertake research, hampering theorists’ ability to acknowledge the true complexity of their target systems and appreciate the place of neuroscience and psychology within the life sciences more generally. My resistance to the project of naturalizing intentionality is that it smacks of the same kind of imperialism whereby special sciences earn the right to their terms by showing how they fit into the framework of a lower level, supposedly more fundamental science. Because metaphysical commitment to neuro/cognitive-representations de facto means being part of the naturalizing project, I’m inviting mind/brain scientists to use their terms autonomously, in a way that is justified by their own lights and explanatory purposes. That said, there is a way to be metaphysically committal about neuro/cognitive-representations which is congenial to my background agenda of greater autonomy for the mind/brain sciences: treat representational relationships as explanatorily basic, emergent, if you like, and don’t try to naturalise them by fitting them into the scheme of information theory plus evolutionary biology. More autonomy is aesthetically unpleasing to people who seek clean unification across the sciences. Fundamentally, I think our world is too complex for this to be a success without gross distortion of reality. At heart, I am a realist!
Thank you for your thoughtful reply, Mazviita! I agree that cognitive science does not need legitimisation for representational explanation through reduction to ‘more basic’ sciences. I don’t think though that naturalising representation, at least nowadays, should (always) count as an instance of that sort of project. Informational teleosemantics makes use of posits that come from the cognitive sciences themselves as well as from areas of biology other than evolutionary theory, such as neural selection and pruning, and developmental and learning processes. So I think that the physics-based model has luckily lost much of its grip.
More generally, I think that the fact that cognitive representations do not need to be naturalised for them to be bona fide scientific posits does not mean that they cannot. It may be worthwhile to see how representation fits with other successful scientific posits, what relations there are between them, etc, independent from any sort of reductionist imperialism. I also think that gross distortions of reality are par for the course in science: representation itself, I’ve argued, is a posit that doesn’t quite capture anything with just those features out there in the world, being rather a highly idealised and abstracted posit. Maybe this is a difference in perspective more than a substantial disagreement: I think scientific success involves (and requires!) such distortions, given the complexity of the world and cognitive and pragmatic constraints. So success and distortion (even gross ones) need not be in opposition.