Commentary on Favela and Machery, “Investigating the concept of representation in the neural and psychological sciences”

Ben Baker

I admit I was somewhat put off by the first two sentences of Favela and Machery’s paper. It feels odd to use “mainstream___ism” to label merely the wide application of a concept – or even just use of the word “representation” – when we are not sure about what theoretical commitments are involved. The authors do quickly go on to illustrate the broad range of phenomena that representations are supposed to help account for across multiple disciplines; attitudes, concept learning, imagery, non-human intelligence, artificial intelligence, and of course how the brain works. They also demonstrate heterogeneity in the ways different researchers have said representations are involved in explanations of cognition. So I might prefer to describe their main question as being whether there is indeed a coherent “ism” in the widespread use of this concept. That said, the authors succeed at shedding new light on this question, helping us see the size and shape of the ambiguity of the key concept. While the paper does not directly answer whether representation is “understood precisely enough to guide the development of hypotheses,” it positions us to develop an answer in discourse, perhaps even to change the answer.

The authors use a clever elicitation study design with carefully chosen contrast words for “representing,” like “carrying information,” “processing,” and “identifying,” providing a window into the precise (or imprecise) use of the concept. I might have added “detect” to the list of contrast terms, but their choices span the relevant conceptual territory. The authors also might have said more regarding the findings in the rare bit of relevant prior empirical work in this vein (Vilarroya, 2017), but they do go well beyond that work by surveying the concept’s implicit entailments among current researchers in multiple fields.

Together the authors’ four studies offer compelling evidence that neuroscientists and psychologists (more than philosophers) tend to be primarily concerned with internal states that depend causally on external stimuli, and perhaps that relate causally to downstream brain processes and behavior, but not with the connotation of intentionality. This comes out of a relatively lack of agreement as to whether a paradigmatic bit of brain activity that correlates with certain stimuli “is about,” “identifies,” or “represents” those stimuli – all terms that are supposed to involve an intentional relation. This is also brought out by Study 4, which found neuroscientists and psychologists to be less willing to attribute error to brain activity. The possibility of error is, if you like, the mark of intentionality, and the results suggest non-philosophers tend not to ascribe this possibility to the neural activity in question. But how do the subjects use the term under investigation?

Crucially, the scattered answers centering around “neither agree nor disagree” suggest that there is general uncertainty around the concept of representation. What are we to do with this? Can we reform the concept? Should we eliminate it?

To work toward an answer to this, I would ask for further discourse around our understanding of “function,” and what might have come out if we elaborated on Study 3. This study manipulated whether the brain activity in question is used by a broader neural network, or whether it is merely found to correlate with the stimulus. As the authors point out, this implicates a “causal role” notion of function (as in Cummins, 1975), but not a teleological notion of function (as in Millikan, 1989, though for my money Ernest Nagel’s pair of papers from 1977 is excellent on this). But these ideas of function can be blurred somewhat, especially in the brief instructions that come with survey questions. The bigger and more well-studied is the network of brain areas in which the given neural activity is integrated, the more likely one is to assume that the network, and thus the activity, contributes importantly to evolution. We can imagine a Study 3A, where the neural activity has a causal role in a smaller, species-specific brain process, and a 3B where it is known to function in a large and evolutionarily old process. Thus, without explicitly bringing in the notion of error and misrepresentation, we could probe the extent to which scientists are concerned with states that implicate the normativity of survive-and-reproduce, and how that plays into their understanding of “representation.”

One interpretation of Favela and Machery’s results – a pessimistic one by my lights, but I suppose I am a philosopher first – is that scientists tend to see content and meaning in strictly causal terms, and are uncertain or outright skeptical of intentionality. The fact that neuroscientists and psychologists more so agreed that the specified brain activity “is about” the stimulus than they did that it “represents” the stimulus makes “being about” is closer to “processing” and “responding to” in this respect. If intentional, potentially erroneous content in the brain is not something science thinks or cares about, then it would be best to eliminate representation-talk from the research; if the “aboutness” of smoke vis-a-vis fire is as deep as it gets, “representation” is just distracting in the way it can import connotations from artistic, political, and other contexts.

But I am a hopeful reformer. I think what this paper brings to light is that we need more discussion across disciplines on how trajectories associated with evolution, development, and learning matter in determining what any given brain activity is about. Causal descriptions only take us so far toward the explanatory aims of psychology and neuroscience, which include the foibles and fantasies of thinking beings. Wherever the current murkiness around intentional ascription comes from – perhaps old ambitions to analytically deduce cognitive phenomena from norm-free “atoms” of neural activity, perhaps a hesitancy to draw conclusions about meanings one might think belong in a humanities department – I think it can and should be clarified. We can, through discourse around teleology and misrepresentation, save representational language for the cases where evidence or shared assumptions are sufficient to support a concept thicker than “causal role,” imbued with the intentionality in terms of which we understand complex and intelligent behavior.

Back to Top