The Fallacy of the Brain as a Faithful Window on the World


The Fallacy of the Brain as a Faithful Window on the World - The Brain is a Predictor of Meaningful Events

 

Today, I received a notice of a talk to be given at Caltech by Romi Nijhawan, Associate Professor, Department of Psychology, University of Sussex Falmer, UK. He entitled his talk Visual Prediction, and the abstract reads as follows:

 

As change in the physical state of a biologically relevant property occurs over time, neural delays should cause an ‘error’ between the instantaneous state ‘registered’ by the central nervous system (CNS) and the ‘physical’ state. Without compensation for such ‘errors’ animals would not survive, as CNS action would be based on outdated information. Contrary to the ‘standard view’ that only motor planning systems compensate for delays, a parsimonious approach is developed which suggests that compensation is a general property of the nervous system. Although motor planning is highly flexible in accounting for the variability in delays (e.g. caused by changes in levels of luminance-contrast, muscle fatigue etc.), it is argued that visual mechanisms also contribute to compensation for delays via a process called Visual Prediction.

 

Yet, while the conclusions stated here seem reasonable, the premise seems to me misguided. The brain knows nothing about the 'physical' state of the world other than what neurons tell it. As it does not know it, it does not try to match the 'registered' state to it or compensate for 'errors'. What the brain does is use the information contained in neuronal states to predict biologically meaningful events. This may be clearest with an example: let's say the appearance of a ball at a certain distance from your face rapidly approaching you causes, in turn, some tens or hundreds of milliseconds later, the percept of such ball, which in turn has been learned to predict an intense pain following it (for the purposes of discussion, let's assume that impact follows the original appearance by 400 ms and your percept by 500 ms). I see no reason to assume that the brain will reconstruct the past and then predict forward from that again. Instead, one would assume the brain will learn that this sight predicts pain within 400 ms, and send motor commands accordingly. Thus, the problem of 'compensating for errors' derives from the illusion that the brain faithfully represents the world. What the brain does is build representations that are useful to convert patterns of neural activity into behavior. Behavior whose usefulness is ultimately measured in terms of fitness.