Sackler Centre for Consciousness Science

Prediction and Expectation

Although it feels like we see the visual world as it objectively is, in fact our brains have to construct it. What is becoming increasingly apparent is that this process of inferring our visual surroundings is predictive: what we expect to see in the present context will guide what we actually see. In the video below is a rotating mask that illustrates this point. For half of the time we are presented with the mask from behind, so that it appears concave. However that is not what we see: we see a strange switch of direction, but the face remains convex. This illusion happens because our brain will not infer the extremely unlikely case of a convex face.


This ‘predictive processing’ account of perception has become increasingly powerful in explaining many of the observations about human experience and behaviour. This approach suggests that the expectations, or predictions, are what determines the contents of perception – not simply the ‘bottom-up’ processing of sensory signals. With a large body of evidence from animal studies supporting the view that neural systems are performing probabilistic inference and computation on incoming sensory information.

One aspect of this sub-groups research within the Sackler Centre focuses on sensory predictive processing and expectations and investigates its relationship to conscious perception.

In the past we have used motion-induced blindness illusion to induce a temporal blindness of a visual target. We found that providing predictive colour information of the target can facilitate its return to conscious perception, suggesting that there is a direct relationship between predictive processes and conscious access to visual information. In a recent study using EEG, we found that the human visual system is capable of learning complex luminance dynamics within the environment. We found that the visual system can track rapid non-periodic luminance sequences, even when the luminance values in a sequence are altered 160 times a second. When the luminance sequence was later repeated, the ‘echo-like’ neural response recorded from occipital sensors became stronger, suggesting both recognition and regularity learning of the sequences had occurred within the visual cortex. We also found that this tracking system was entirely unconscious, suggesting that predictive processes can exist without consciousness involvement.


In a recent study we investigated how this predictive process might occur. We measured electrical activity over the scalp, and looked at neural oscillations (thought to instantiate communication across brain areas) over occipital cortex. This region of the brain is specialised for vision. We found that the propensity to incorporate expectations into the construction of our visual world is not constant over time. Instead, it fluctuates or cycles approximately every 100ms, following the shape of these oscillations.

This result suggests that the extent to which the brain will use expectations in perception is shaped by the state of visual brain regions before we have even seen anything: sometimes the brain is ‘more prepared’ to base its inferences on sensory signals, while at others it is ‘more prepared’ to base its inferences on expectations. It might be that this preparedness reflects those expectations being communicated to visual regions from other parts of the brain. Alternatively, it might reflect a compensatory mechanism, whereby the visual regions recruit expectations for its processing more when it is in a ‘suboptimal’ state for processing sensory signals. In both cases, this work tells us that neuroscience theories attempting to explain how the brain uses expectations need to account for this finding that expectations are not available or used continuously, but rhythmically.