Testing Predictive Processing with MEG and Eye-tracking

Visual perception is not what it seems to be. We usually have the impression that the surrounding is present in a detailed, stable and continuous way. A closer look, however, reveals that visual acuity is only good within a very small region of the visual field (fovea) and that we make extremely fast, jumping eye-movements, so-called saccades, about three times a second. These eye-movements drastically change the visual input and even render us momentarily blind. Surprisingly, we are not aware of these disruptions at all. One hallmark explanation for the experience of visual stability despite saccades is the idea that, before an eye-movement, a copy of the motor command for the eye-movement is sent to visual brain areas, the so-called efference copy. This copy allows sensory brain areas to create a prediction, based on peripheral low-acuity visual input combined with expectations, to anticipate what will be perceived in detail after the eye-movement. This predicted visual input is supposed to supersede, at least in parts, the actual visual input during the saccade. The idea was first expressed more than 150 years ago and since then many researchers have provided evidence for predictive processes around the time of saccadic eye-movements. Not only in the area of vision, theories of predictive processes have flourished in the last decades. Among the most influential ones is the concept of predictive coding, which aims at an explanation of brain function in general. Crucially, this framework postulates separate top- down (prediction) and bottom-up (prediction error) processes among brain networks. The aim of the present project is to find out whether predictive processes around the time of saccadic eye-movements can be explained within the general framework of prediction and prediction error processes. For this goal, I will conduct a combined neuroimaging (magnetoencephalography, MEG) and eye-tracking study and apply information transfer measures on the resulting brain data. This project has, thus, the potential to pin down a neural correlate of the efference copy and provide an explanation for the subjective impression of visual stability in terms of predictive coding.

Within about the last two decades, the topic of predictive processing has become one of the major themes in the cognitive sciences. In particular within the cognitive neurosciences, the main principle of brain function has been considered a predictive principle according to which the brain constantly tries to predict upcoming input based on bottom-up sensory input and top-down signals derived from learnt associations. The origins of this idea can be traced back more than hundred years and it has indeed been extremely successful in more recent time in explaining many phenomena in the areas of basic sensorimotor processing and even higher-level cognitive functions. It has also been argued that the principle of predictive processing can explain one of the main mysteries of visual perception, that is, it could answer the question why our perception of the world appears to be stable despite the highly dynamic nature of natural visual input and the comparatively limited information that we have at each moment in time. Humans make about three to four eye movements per second, and with each eye-movement, visual input changes drastically and abruptly. Still, we are not aware of this change. Some predictive mechanisms have already been proposed to explain this apparent discrepancy. The idea of the present project was to test whether these mechanisms can be conceived of as prediction and prediction-error processing within a more general account of predictive processing. We tested this predictive processing account in neurocognitive experiments with human participants in which brain activity was recorded simultaneously with gaze behaviour and we investigated how brain activity and gaze behaviour changed when the regularities and the structure in the visual input changed with respect to their predictability. This method allowed us to determine a set of neurophysiological mechanisms that are active around the time of eye-movements and that indicate information processing in a specific way that is indicative of prediction and prediction-error processing. However, some of these mechanisms turned out to be less related to predictive processing than expected. It became clear in the course of the project, that certain types of non-predictive adaptation could provide a good explanation for some of the mechanisms that we had identified. We, thus, concluded that predictive processing can only account for parts of the mechanisms that are responsible for why we perceive a stable world despite the inherent dynamics and incompleteness of actual visual input.