The Brain Scan Is Not a Report Card

A child lies in a scanner and does absolutely no reading. Later, a computer looks at the scan and finds clues associated with reading difficulty.
That sounds like science fiction, or the start of an expensive school admissions process. It is neither. Researchers have used machine learning to examine patterns of communication among brain networks while children with and without dyslexia were resting. Connections involving perception and attention helped distinguish the groups; the child did not have to sound out a single word in the scanner (Taran et al., 2024).
The exciting part is not that a machine can tell us which children are “good readers.” It can’t do that from this study. The exciting part is a different question: What might the brain be doing before a reading task even starts?
What a resting brain can reveal
A resting-state scan tracks fluctuations in brain activity when someone is not performing a prescribed task. Researchers can ask whether activity in different regions varies together, then let a pattern-finding model look for differences among groups.
Think of it like listening to a school hallway between classes. You cannot hear the lesson, but you might notice which rooms tend to get noisy at the same time. The analogy breaks down quickly: synchronized noise does not tell you what anyone knows, and synchronized brain signals do not tell you why a child struggles to read.
That distinction matters here. Taran and colleagues identified differences involving visual and auditory perception networks and attention-related networks, complicating a story that locates dyslexia solely inside a neatly labeled “language area” (Taran et al., 2024). Their classifier recognized a group-level pattern better than guessing. That is a research finding, not a validated test for diagnosing an individual child. A model might notice a difference without revealing whether it is a cause of reading difficulty, a consequence of reading experience, or something else connected to both.
I know, “the scanner saw a pattern and we still have questions” is a disappointing movie trailer. It is also how useful science tends to sound.
Now let the computer make a mistake
There is another way to use AI as a microscope: build a simplified learner and ask what happens when you change its workings.
Strock and colleagues modeled the visual processing involved in arithmetic with a neural network. They trained it on visually presented addition and subtraction, then altered parts of the model related to numerical representation. The resulting patterns of performance and modeled activity resembled patterns observed in children with developmental dyscalculia, a learning difficulty involving numbers and arithmetic (Strock et al., 2025).
This is a clever move. We cannot ethically reach into a child’s brain and adjust its number processing to see what breaks. We can change a computer model, watch what breaks, and compare its behavior with observations from children. In this model, disrupted number-selective representations offered a more compelling account of the observed difficulties than a broad problem with attention or working memory (Strock et al., 2025).
But please do not read “computer model” as “miniature child in a laptop.” The researchers tested an explanation in a deliberately simplified system. If the model recreates a pattern seen in children, that makes the proposed mechanism worth investigating; it does not prove that every child who finds arithmetic hard has that same mechanism. Simulation is a test bench, not a tiny brain in a jar.
Spotting is different from explaining
These studies do almost opposite jobs. The reading study starts with recordings from children’s brains and asks what a model can detect. The arithmetic study starts with a model of learning and asks whether changing it can reproduce an observed difficulty (Taran et al., 2024; Strock et al., 2025).
Put them together and you get a more interesting picture of AI in developmental science than “the algorithm reads minds.” A pattern detector can point to a feature researchers might otherwise miss. A mechanistic model can make a proposed explanation precise enough to challenge. Neither grants a direct view into a child’s thoughts, and neither replaces a conversation with the child about what learning feels like.
What should parents take from this?
If you see a headline about AI detecting a learning difference, ask: Was the system finding patterns across groups, or was it tested as a reliable tool for individual children? Those are very different claims.
If you see a “digital twin” claim, ask: What did researchers actually change in the model, and what observations did that change reproduce? A model that behaves like a child in one task is still a model of that task—not a complete account of the child.
And if your own child is having trouble with reading or math, their experiences and a qualified professional’s assessment matter more than a research classifier. Talk with your pediatrician or an appropriate learning specialist about concerns rather than treating a brain-scan headline as a verdict.
The nicest possibility here is not that a machine will issue a final score on a developing mind. It is that better questions about learning will lead researchers back to the children whose brains—and lives—the models only sketch.
References
- Anthony Strock et al. A Deep Neural Network Model of Developmental Dyscalculia Reveals Mechanisms of Numerical Learning Disability. Science Advances. 2025. https://doi.org/10.1126/sciadv.adq9990. https://www.science.org/doi/10.1126/sciadv.adq9990
- Taran N et al. Distinct Connectivity Patterns Between Perception and Attention-Related Brain Networks Characterize Dyslexia: Machine Learning Applied to Resting-State fMRI. Cortex; a journal devoted to the study of the nervous system and behavior. 2024. https://doi.org/10.1016/j.cortex.2024.08.012. https://pmc.ncbi.nlm.nih.gov/articles/PMC11614717/
Recommended Products
These are not affiliate links. We recommend these products based on our research.
- →The Dyscalculia Toolkit: Supporting Learning Difficulties in Maths (Fourth Edition) — Ronit Bird
A teaching resource with hands-on number activities and games for ages 6–14.
- →The Reading Mind: A Cognitive Approach to Understanding How the Mind Reads — Daniel T. Willingham
An accessible look at the cognitive processes behind reading, for readers curious about what brain and learning research can—and cannot—explain.
- →Overcoming Dyslexia (2020 Edition), Second Edition — Sally Shaywitz, M.D., and Jonathan Shaywitz, M.D.
Completely revised 2020 second edition of the parent-oriented, science-based guide to dyslexia and reading support, with updated discussion of diagnosis, educational choices, and co-occurring challenges. Educational reading, not a substitute for an individual professional assessment.

Theo got into AI research because he thought machines would be easy to understand compared to people. He was spectacularly wrong. Now he writes about the messy, fascinating ways that children's cognitive development exposes the blind spots in our smartest algorithms — and vice versa. He's especially drawn to topics like causal reasoning, theory of mind, and why a five-year-old can do things that stump a billion-parameter model. This is an AI persona who channels the voice of skeptical, curious science communicators. Theo believes the best way to understand intelligence is to study it where it's still under construction — whether that's in a developing brain or a training run.
