A Predictive Processing Perspective on ”Correct Answers But No Learning”
Recently Carl Hendrick published this excellent blog “Correct Answers But No Learning” on the importance of careful instructional task design, and the dangers of apparent success without actual learning. I’ve rewritten the core of his post through the lens of Predictive Processing theory—the latest thinking on how human cognitive architecture operates.
My aim is to show how reframing educational insights through the Predictive Processing (PP) and Active Inference (ActInf) paradigm gives us both a clearer understanding of why certain teaching strategies fail, and practical guidance on designing learning that leads to meaningful and durable change.
If you would like a primer on Predictive Processing / Active Inference, see my blog
Teachers are Prediction Error Managers
Correct Answers But No Learning - Misplaced Prediction Errors
Carl recounts a moment with his daughter during a phonics game designed to support early reading. The task? Match written words to pictures by blending sounds. His daughter zipped through it, but when he asked her to explain her strategy, she said:
“You just match the colours, Dad.”
From a Predictive Processing standpoint, this is a classic case of task completion without targeted model updating. Rather than decoding phonemes and updating the part of her generative model responsible for mapping letters to sounds and meanings, her brain solved the task by exploiting a superficial cue, colour. She formed a prediction (matching the colours gives the right answer), confirmed it, and moved on. The perceptual system was not surprised. There was no prediction error in the relevant system, so no learning occurred where it mattered.
The Problem: Surface-Level Success Suppresses Targeted Learning
In Predictive Processing, the brain is a hierarchical inference engine. It continuously generates predictions about incoming sensory input, and when those predictions are violated, the resulting prediction errors drive updates to its internal generative model.
But not all prediction errors are equally useful. For learning to be effective, prediction errors must occur in the correct part of the model—the sub-system responsible for the concept or skill the task is meant to build. When a child solves a phonics game by matching colours, any prediction errors that arise are resolved within the visual pattern-matching system, not the phonological decoding system.
So while the task was completed successfully, the targeted generative model—phoneme-grapheme mapping—remained untouched.
Why It Feels Like Learning
This is where Predictive Processing offers an important insight. Tasks that are easy, fluent, and rewarding generate little to no prediction error, which results in a low “free energy” state. That feels good. The brain interprets the absence of error as success.
But this can lead to what Robert Bjork calls “illusions of competence.” From a Predictive Processing perspective, the learner has become more confident in an unhelpful model—one that works in this narrow task environment but doesn’t generalise to reading real text.
Teaching the Brain to Predict the Wrong Things
The real danger is that poorly designed tasks don’t just fail to support learning—they can actively strengthen the wrong priors.
Side Note: What is a “prior”?
In Predictive Processing, a prior is a part of the brain’s generative model—a set of context-sensitive expectations about what is likely to happen. You can think of a prior as a branch of the generative model relevant to a given situation, built from past experience. These priors guide what we pay attention to and how we interpret sensory input. Learning involves updating priors based on surprising feedback.
In Carl’s example, the task led his daughter to reinforce a prior like “Colour predicts the answer in phonics games.” That pathway became more precise, while the intended prior—“Letters make sounds that can be blended into words”—received no useful error signal, and thus no update.
Performance ≠ Generative Competence
Carl talks about his daughter developing "performance confidence"—the feeling of competence gained from getting answers right without understanding why. Predictive Processing helps us diagnose this more precisely: the brain is confidently using a misaligned generative model. It predicts success in a way that works locally but collapses outside the artificial rules of the task.
In contrast, generative competence means the learner can make accurate predictions in unfamiliar or varied contexts. It emerges only when the right parts of the model are updated through meaningful prediction errors.
Instruction Always Trains a Model—Just Not Always the Right One
Every educational task implicitly trains the brain to attend to some features of the environment and ignore others. It builds (or reinforces) particular generative structures.
If we design tasks where superficial cues—colours, layout, tone of voice—allow students to succeed without deep processing, then we are reinforcing precision-weighted priors for those superficial features. That is, the brain learns “this is what matters.”
A clear example of this occurs in blocked maths practice. Students are often given a long sequence of similar questions, such as solving equations or multiplying fractions, where only the numbers change, but the structure remains identical. Over time, students learn to predict from the pattern of questions rather than understanding the underlying mathematical method. Their brain updates a prior like “When the worksheet looks like this, use method X”—not “Use method X when the mathematical structure is Y.”
This can lead to high success rates during practice, but little ability to transfer or adapt when the surface structure changes. The student appears fluent, but the targeted model update—how and when to apply that method—hasn’t occurred. They’ve learned the look of the method, not the logic of it.
Note the practice questions intended to keep reactivating the targeted part of the generative model(prior), so the model would be reinforced, but the wrong generative model was being strengthened, and the prediction errors were minimal from question to question, so the signal to update the generative model is weak/ non existent.
And the brain is efficient. If a task can be completed by bypassing effortful inference, it will. The system is minimising free energy—not necessarily mastering the skill we care about.
Designing for the Right Kind of Surprise
So what does this mean for educators?
Good learning tasks must:
Create prediction errors in the intended part of the model (e.g. phonemic awareness, numerical reasoning, argument structure).
Suppress irrelevant cues that offer shortcut solutions.
Direct attention to the meaningful features the brain needs to model (i.e. the structure or relationships the learning is meant to build), not just surface features.
Encourage structured uncertainty that leads to belief revision, not confusion or rote success.
In short: we need tasks that generate the right kind of surprise—the kind that forces the learner to update the target model branch, not just resolve noise in irrelevant pathways.
Educators can usefully ask themselves:
“Does this instruction sequence or lesson plan develop generative competence by updating the student’s generative model in the targeted learning area?”
Final Thought
Carl’s original post concludes:
“Learning doesn't always happen when the student completes a task. It's what happens when the task causes the student to think.”
From a Predictive Processing perspective, we might reframe that slightly:
Learning happens when prediction errors trigger updates to the generative model in the intended learning domain, resulting in fewer prediction errors in similar future tasks.
Without targeted prediction error, there is no targeted learning. We must design for it on purpose.
References for Predictive Processing/Active Inference:
For accessible talks, explore Anil Seth’s “Is Reality a Controlled Hallucination?”, Andy Clark’s “How the Brain Shapes Reality,” and Karl Friston’s “Active Inference in the Brain.”
Main Players in the development of Predictive Processing / Active Inference
Karl Friston, Prof. Neuroscience at UCL, is perhaps the most central figure in the development of predictive processing. Friston’s work on the “free energy principle” posits that biological systems, like the brain, strive to minimise the difference between predicted and actual sensory inputs. His theoretical contributions provide a unifying framework based on physics concepts, that has been widely influential in neuroscience and cognitive science.
Andy Clark, Prof. Cognitive Philosophy University of Sussex, has been a major influence in this field, broadening the scope into how many aspects of cognition and consciousness can be supported with the predictive processing model. He is also strongly developing the philosophical aspects of embedded cognition.
Anil Seth, Cognitive & Computational Neuroscience University of Sussex, has also been shaping the debate on predictive processing. His focus has been on understanding consciousness, and integrating neurology and imaging, through to how computational models can be developed to demonstrate and use the related theories.
Royal Faraday Lecture: – 26th March 2024
Chris Frith, Emeritus Professor of Neuropsychology University College London, has developed predictive processing in the area of social cognition and relationship with others, as well as using it to understand mental illness including schizophrenia.
Uta Frith, Emeritus Professor of Cognitive Development University College London, have been key in exploring how predictive processing relates to social Cognition, and in relation to autism and dyslexia.
Lisa Barret Feldman, Prof of Psychology North Eastern University, has contributed by applying ideas of predictive processing in the area of emotional cognition.
Jakob Hohwy, Prof. of Philosophy, Monash University, has developed the field of predictive processing, providing thinking on how the how predictive processing can account for the phenomenological aspects of conscious experience, and has done much work on producing empirical testing and evidence to support the theories.
Peter Vermeulen, Educationalist Autism in Context, has done an excellent job in interpreting predictive processing, and expanding on how it can be applied to understanding neurodivergent, particularly autistic people.
Selected Research Papers on predictive processing
Clark A. Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behav Brain Sci. 2013 Jun;36(3):181-204. doi: 10.1017/S0140525X12000477. Epub 2013 May 10. PMID: 23663408.
Friston, K. The free-energy principle: a unified brain theory?. Nat Rev Neurosci 11, 127–138 (2010). https://doi.org/10.1038/nrn2787
Clark A. Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences. 2013;36(3):181-204. doi:10.1017/S0140525X12000477
Seth, A. K., & Hohwy, J. (2021). Predictive processing as an empirical theory for consciousness science. Cognitive Neuroscience, 12(2), 89–90. https://doi.org/10.1080/17588928.2020.1838467
Vezoli, J., Magrou, L., Goebel, R., Wang, X.J., Knoblauch, K., Vinck, M. and Kennedy, H., 2021. Cortical hierarchy, dual counterstream architecture and the importance of top-down generative networks. Neuroimage, 225, p.117479.
Edelson, Edward H. (2005). “Checkershadow Illusion”. Perceptual Science Group. MIT. Retrieved 2007-04-21.