Is Consciousness Predictive? Rethinking the Bayesian Brain and Visual Illusions

By Leandro Castelluccio

In recent years, there has been a surge in the application of Bayesian statistical concepts across various disciplines, with neuroscience being no exception. Increasingly, researchers are exploring Bayesian inference as a framework to understand perception — and even consciousness itself. This approach, while elegant and mathematically compelling, may also carry conceptual blind spots that are worth re-examining.

A common entry point into this topic is the analysis of visual illusions — phenomena where what we see appears to diverge significantly from what is actually present in the environment. These illusions are often cited as evidence that our conscious experience is not a direct mirror of reality, but rather a product of the brain’s inferential processes.

Take, for instance, the well-documented rotating hollow mask illusion. Despite being presented with the concave side of a face, most people perceive it as convex. This illusion has been interpreted as evidence that the brain, conditioned by evolutionary and social imperatives, “expects” faces to be outwardly shaped and thus “corrects” the input accordingly. In Bayesian terms, this suggests that what we consciously perceive is not raw sensory data, but the brain’s best guess based on prior experience.

However, this interpretation raises important questions. First, it assumes that perception itself is inherently predictive. But is that always the case? Is the brain truly engaging in an active prediction model at all times, or is there a more fundamental process at play — perhaps the probabilistic activation of neural patterns conditioned by previous inputs, without the need for constant inferential updating?

This distinction becomes more pronounced when we consider the limited scope of perceptual illusions across sensory modalities. Why are such compelling and persistent illusions predominantly visual? If Bayesian inference were a generalized model for all perceptual consciousness, we would expect similarly rich phenomena in modalities like touch, olfaction, or proprioception — but such examples are rare.

Consider another well-known illusion: when two adjacent trains are stationary, and one begins to move, passengers in the still train may momentarily feel as though their train is in motion. Does the brain expect that the self is more likely to move than the environment? The answer is no. The experience is not born of an expectation but a momentary conflict between sensory inputs, such as vestibular and visual cues. Again, this casts doubt on the idea that perception is always filtered through predictive Bayesian priors.

Moreover, if the Bayesian brain hypothesis held in strict terms, one would expect that incorporating new sensory information could correct false inferences. Yet, many visual illusions persist even after prolonged exposure or conscious awareness of their illusory nature. If prediction errors cannot be rectified despite ample sensory data, we may be facing a conceptual boundary — either an unfalsifiable theory or one that lacks sufficient explanatory power for the phenomenology of consciousness.

This leads us to a subtler point: perhaps what we call “prediction” is better understood as statistical predisposition. Certain neural pathways may be more likely to activate in response to familiar stimuli, not because the brain is simulating future input, but because the architecture of perception is historically and biologically shaped to respond in particular ways. In this light, perception conditions perception — but does not necessarily imply a forward-looking, dynamic predictive model.

If visual illusions such as the hollow mask are simply artifacts of how certain stimuli are processed — and not clear evidence of predictive coding — then we must be cautious about using them as proofs of a Bayesian model of consciousness. The lack of equivalent illusions in other sensory modalities further challenges the generalizability of this approach.

In short, while Bayesian inference remains a powerful and fruitful tool for modeling cognition and perception, its extension to consciousness demands careful scrutiny. To conflate statistical models with subjective experience may be to mistake the map for the territory.

References

General Application of Bayesian Inference in Neuroscience:

Friston, K. (2010). The free-energy principle: a unified brain theory?
https://www.nature.com/articles/nrn2787
(A foundational paper linking Bayesian inference and brain function.)

Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science.
https://pubmed.ncbi.nlm.nih.gov/23663408/

Hohwy, J. (2013). The Predictive Mind (book)
https://global.oup.com/academic/product/the-predictive-mind-9780199682737
(Useful for readers wanting a deeper dive into predictive processing theory.)

Rotating Hollow Mask Illusion:

Video demo and explanation of the illusion:

Gregory, R. L. (1997). Knowledge in perception and illusion.

https://pmc.ncbi.nlm.nih.gov/articles/PMC1692018/

Critiques and Limitations of Predictive Coding in Consciousness:

Williams, D. (2028). Predictive Processing and the Representation Wars.
https://pubmed.ncbi.nlm.nih.gov/31258246/

Doerig, A., Schurger, A., & Herzog, M. H. (2021). Hard criteria for empirical theories of consciousness. Cognitive neuroscience, 12(2), 41–62.

https://www.tandfonline.com/doi/full/10.1080/17588928.2020.1772214