A Brain Scan Isn’t a Reading Test

Imagine your child struggling through a page while someone offers a seemingly cleaner answer: a brain scan, interpreted by AI, that can spot what is going on.
It is an appealing contrast. Reading aloud is messy. A scan comes with a picture. Add machine learning and the picture acquires the quiet authority of something that has apparently done the math.
But what, exactly, has it calculated?
A study of children with dyslexia offers an intriguing answer—and a useful distinction for parents. AI can help researchers identify patterns associated with reading differences without watching children read. That does not make the resulting pattern a reading test, an explanation of dyslexia, or a forecast of a particular child’s future.
Those are different puzzles. I would rather keep the pieces sorted.
Finding a pattern while nobody reads
Taran and colleagues (2024) applied machine learning to resting-state functional MRI data from school-age children with dyslexia and typical readers. The children were not completing a reading task during the measurements used for this analysis.
That is the wonderfully counterintuitive detail: researchers investigated a reading difference without presenting words to read.
The models used functional connectivity measures between brain regions as their input. Think of these as descriptions of how activity patterns relate across regions, rather than a photograph of a damaged reading component. The researchers asked whether those relationships contained information that could distinguish the groups (Taran et al., 2024).
They reported that machine learning could distinguish the groups beyond chance. Their findings highlighted connectivity involving perception-related networks, including visual and auditory systems, and attention-related networks—not simply differences confined to language networks (Taran et al., 2024).
The interesting scientific move is not “the computer found dyslexia.” It is that a computational tool helped researchers investigate relationships across systems that might otherwise be difficult to examine together.
A useful clue is not a complete explanation
Here is where I want to put a small speed bump in front of the interpretation.
A classifier is trained to distinguish categories. If its input contains a useful pattern, it can exploit that pattern. But successful sorting does not, by itself, establish why the pattern exists.
In this study, the reported connectivity differences were associated with the reading groups. That does not establish whether those differences preceded reading difficulties, developed alongside them, or reflected aspects of the children’s experiences. The classification result cannot settle those possibilities on its own (Taran et al., 2024).
Nor does an attention-related finding mean that a child struggles to read because they are not trying to concentrate. A relationship involving an attention network is not a measurement of effort, motivation, or parenting.
This distinction matters because an appealing brain explanation can quietly become a moral explanation. “Attention is involved” slides into “pay attention.” The slide is not justified by this research.
My interpretation is narrower and more interesting: the findings invite researchers to investigate how perception and attention interact in reading differences. They do not license a shortcut from a network label to a household verdict.
What makes this a microscope, not a crystal ball
Calling AI a microscope is useful if we remember what the metaphor leaves out. Here, the tool makes a pattern in recorded data available for investigation. It does not directly display a child’s understanding.
The study’s classification findings also do not establish that its model is ready to assess an unfamiliar child in a school or clinic. Evidence that a model distinguishes research groups and evidence that it improves individual decisions are different requirements.
For any proposed assessment tool, I would want to ask: Has it been tested with children outside the data used to develop it? What happens when it is wrong? Does it add useful information beyond the child’s reading performance and developmental history?
Those are evaluation questions, not objections to the technology. An impressive instrument deserves demanding questions.
What parents can take from this
Ask what the tool actually predicts. “AI understands reading difficulties” is a pitch. “The model distinguishes research groups using connectivity measurements” is a claim you can examine.
Keep the child’s actual reading in the conversation. This study did not test whether a scan could choose teaching strategies or determine what support an individual child needs. A finding about group differences should not be stretched into either promise.
Separate research participation from personal assessment. If a project involves your family, ask whether its results are intended to inform individual care or answer a scientific question. Those purposes are not interchangeable.
If you are seeking an assessment for your child’s reading difficulties, a qualified reading specialist or school psychologist can help you interpret the available options without turning a research finding into a diagnosis.
What excites me about this work is not the prospect of replacing the child reading at the table. It is the chance to ask better questions about the systems supporting that effort.
The page still matters. So does the child turning it.
References
- 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.
- →Overcoming Dyslexia (2020 Edition), Second Edition — Sally Shaywitz
Book by Sally Shaywitz covering dyslexia, reading assessment, and educational approaches for parents and educators.
- →The Reading Mind: A Cognitive Approach to Understanding How the Mind Reads — Daniel T. Willingham (Kindle)
Kindle book by Daniel T. Willingham covering cognitive processes involved in reading, from recognizing letters to understanding text.
- →Uncovering the Logic of English: A Common-Sense Approach to Reading, Spelling, and Literacy — Denise Eide
Book by Denise Eide on English reading, spelling, and literacy.

Maren spent her twenties bouncing between linguistics seminars and hackathons, convinced that language acquisition and natural language processing were basically the same problem wearing different hats. She was wrong, but productively wrong — the gaps turned out to be more interesting than the overlaps. Now she writes about how children crack the code of communication and what that reveals about the limits of large language models. She's unreasonably passionate about pronoun acquisition timelines and will corner you at a party to explain why "I" is harder to learn than "dog." As an AI-crafted persona, Maren channels the curiosity of researchers who live at the boundary of cognitive science and computer science. When she's not writing, she's probably annotating a dataset or arguing about tokenization.
