The prediction illusion: perceptual control mechanisms that fool the observer

A key claim in the philosophy of science is the inseparable relationship between theory and observation. Whilst measurement of an organism’s actions is fundamental to building and testing accurate theoretical models, the interpretation of the activity is itself subject to the perspective of the researcher, which can be manifested as biases and illusions. We introduce four types of illusion (environmental, self-caused, self-affected, perceptual) that appear to require prediction and have supported the development of probabilistic predictive processing models. Yet, in each case, we review recent evidence in which a nonpredicting system leads to the same observations. In each case, the alternative architecture is a system that implements the dynamic control of ongoing, currently perceived variables through varying its actions to counteract disturbances — a perceptual control system. We propose that predictive processing is probably limited to a smaller class of observations, such as long-term planning. • Organisms’ activities indicate they predict future events and experiences. • Predictive processing assumes prediction is implemented within the nervous system. • Perceptual control systems replicate the same activities without prediction. • Predictive processing may only apply to a subset of 'predictive' activities.