Metaphors in neuroscience (III) - Neural representations

A commonplace in neuroscience is to say that the brain represents the world. In particular, sensory systems are thought to represent images or sounds. A representation is an object, for example a picture or a sculpture, that is meant to be recognized as another object. Both the original object and its representation are to be presented to the same perceptual system and recognized by that system as perceptually similar. The idea that the perceptual system itself might represent an external object seems quite peculiar: it seems to entail that there is a second perceptual system that sees both the external object and the representation made by the first perceptual system. The metaphor cannot apply in this way. But then what do people mean when they say that the brain represents the world? A clue might be provided by this quote from David Marr's book “Vision” (1982):

If we are capable of knowing what is where in the world, our brains must somehow be capable of representing this information.”

Here “knowing” cannot simply mean acting as a function of the external world, because it has been well argued that representations are not necessary for that – simple control systems can be sufficient (see e.g. Rodney Brooks and Braitenberg's vehicles). Thus “knowing” must be meant in a stronger sense, the ability of manipulating concepts and relating them to other concepts. If it is assumed that anything mental is produced by the brain, then somehow those concepts must be grounded in the activity of neurons. In what sense does that activity form a “representation” of external objects? For this metaphor to make sense, there must be a system that sees both the original object and its mental representation, where the representation is an object that can be manipulated. The possibility of mental manipulation entails working memory. So for the brain, representing the world means producing persistent neural activity, structured in such a way that it can be compared with the direct sensory flow coming from the external world.

What is surprising is that Marr and many others do not generally use the representational metaphor in this proper way. Instead, the activity of sensory systems, for example the primary visual cortex, is seen as representing the external world, in the same way as a photograph represents the external world. But unless it is meant that the rest of the brain sees and compares retinal activity and cortical activity, cortical activity is a presentation, not a representation. It might be a representation for an external observer (the one that sees both the world and the photographs), but not for the brain. Thus the metaphorical idea that mental/neural representations mediate perception is somewhat self-contradictory, but unfortunately it is one of the dominant metaphors in neuroscience.

P.S.: see this later comment

2 réflexions au sujet de « Metaphors in neuroscience (III) - Neural representations »

  1. I think presentation is only a better metaphor if insofar as (say) V1 activity is a faithful copy of the external environment. However, attentional modulation of V1 activity would suggest that this assumption is incorrect, and in instead it seems to be the case that even during low-level perception control processes manipulate these representations.

    I think a better way to think about representation is how faithfully activity reflects the external environment. When bipolar cells sum the activity of several photoreceptors, the activity becomes less similar to the environment, and instead it is made easier to find patterns. This happens at many level of perception: signal are filtered in order to more easily find the commonalities that will determine a motor response. Abilities like speech perception would be impossible without representation, as each instantiation of a word is unique (noisey). In terms of comparing our representations against what they actually stand for (ie. the environment), we are also constantly generating prediction memories (undeniably representations, as they have no immediate sensory basis) which we compare against our representation of outcomes.

    Representations are useful ways to describe the information-reduction processes that allow for control systems to employ general algorithms for the selection of motor responses.

  2. Hi Harrison. My point is simply that the term "representation" is a misleading metaphor. Photographing things don't help you at all with the problem of unique instantiations of categories. In fact, it seems to me that what you are describing should be called abstractions, rather than representations.

Laisser un commentaire

Votre adresse de messagerie ne sera pas publiée. Les champs obligatoires sont indiqués avec *