Our guessing brains

Ever misheard a word only to realize a little while later what was said? That minor hiccup may seem random, but it is your brain in action – constantly making educated guesses about the world around you. These “guesses” are not accidental or clumsy but rather powerful, structured processes that lie at the heart of how we see, hear, and understand the world. Welcome to the world of predictive coding. 

In neuropsychology, predictive coding is a theory that explains how the brain processes information by constantly predicting what it expects to perceive. It compares these predictions to incoming sensory data and updates its model of the world based on errors in those predictions. Let’s use Google Maps as an example: When traffic is ahead of you, Google Maps is able to anticipate your path but adjusts when reality does not match the plan. 

How does it work? – Using Marr’s levels 

At its core, predictive coding proposes that the brain is a guessing machine where it builds internal models of the world and continuously refines them by comparing its predictions with actual sensory inputs (what it expects to perceive vs what is perceived). If there’s a mismatch between the two, known as a prediction error, the brain adjusts and modifies its internal model to better align with reality. Over time, this helps us perceive the world more efficiently and respond more accurately to new information. According to Gabhart et al. (2025), these predictions help reduce energetic costs by ensuring the brain is not constantly reinventing the wheel – it guesses first, then checks. 

Predictive coding can be understood on three levels, following the framework proposed by Marr (1982): the computational, algorithmic and implementation levels. Sprevak (2023) outlines that the computational level refers to when the brain tasks itself to minimize prediction errors. The algorithmic level refers to when the prediction units generate expectations while error units signal mismatches between expectation and reality. Lastly, the implementation level refers to the physical reality in the brain. The prediction and error units are thought to lie in the neocortex, with different populations of neurons in areas like the visual or auditory cortex handling the guessing game. 

What does the brain predict and more importantly, how?

With all these levels going on, the question then comes in, do our brains only predict one thing at a time? No, not at all. Our brains are predicting multiple layers of information – from basic sensory input (shapes, sounds etc.) to more abstract concepts like meaning and grammar. Caucheteux et al (2023) found that different brain regions predict over different timescales and levels of abstraction. For example, the brain may anticipate a sentence structure while also predicting the next word. Going back to language processing, higher-level areas of the brain are responsible for long-range predictions (e.g., anticipating the punchline of a joke) whereas lower-level areas handle short-term patterns (e.g., individual word sounds). 

Top-down, bottom-up: Conversations in our brains

One of the most exciting parts of predictive coding is how it reimagines the flow of information in the brain. Previously, perception was seen as largely bottom-up where sensory comes in and the brain processes it step-by-step. However, predictive coding changes this completely. 

With the concept of predictive coding, information flows both ways: top-down and bottom-up. Top-down pathways send predictions to lower brain areas while bottom-up signals carry prediction errors back up the chain. This exchange allows for more dynamic adjustments to take place (Shipp, 2016). For example, you see someone in the crowd that looks like your friend, your brain fills in the gaps. But if you get closer and hear that their voice does not match that of your friend’s, prediction error kicks in and you realize that you are moving towards someone you do not know. 

Is this just a fancy way to say ‘guessing’?

Predictive coding is far more than mere guessing. It is grounded in Bayesian inference – a mathematical framework for combining prior knowledge with new evidence to arrive at the most probable conclusion. The brain will not just randomly choose moments to make a guess, it weighs possibilities, updates beliefs, and constantly learns. 

Millidge et al. (2022) highlights that this theory may even offer explanations for functions around the cortex, drawing connections between neuroscience and machine learning. Some researchers have also compared the brain’s predictive strategy to deep learning models albeit with more biological elegance.

Conclusion

Predictive coding and understanding it matters as it shifts how we think about perception, cognition, and even mental health as a whole where prediction symptoms may be impacted. It is not just a theory about brain functions, it is a lens through which we can better understand learning, adaptation, and consciousness. 

So, is your brain always guessing? Yes, it is but it is guessing wisely and that is what makes it so powerful.

 

Written by:

Cassandra Selvan

Singapore University of Social Sciences

 

References:

Caucheteux, C., Gramfort, A. & King, J. R. (2023). Evidence of predictive coding hierarchy in the human brain listening to speech. Nature Human Behaviour, 7, 430-441. https://doi.org/10.1038/s41562-022-01516-2 

Gabhart, K. M., Xiong, Y. S. & Bastos, A. M. (2025). Predictive coding: A more cognitive process than we thought? Trends in Cognitive Sciences. https://doi.org/10.1016/j.tics.2025.01.012 

Marr, D. (1982). Vision: A computational approach. MIT Press.

Millidge, B., Seth, A. & Buckley, C. L. (2022). Predictive coding: A theoretical and experimental review. Cornell University. https://arxiv.org/abs/2107.12979

Shipp, S. (2016). Neural elements for predictive coding. Frontiers in Psychology, 7. https://doi.org/10.3389/fpsyg.2016.01792 

Sprevak, M. (2023). Predictive coding I: Introduction. Philosophy Compass, 19(1). https://doi.org/10.1111/phc3.12950