AI Brain Successfully Mimics Dyslexia and Spots Fonts That Improve Reading Researchers at EPFL have achieved a significant breakthrough in understanding dyslexia by using advanced artificial intelligence to model the condition in an AI brain. This development opens new pathways for identifying effective interventions, including font types that may improve reading accessibility for individuals with dyslexia. Dyslexia, a common learning disorder affecting reading, spelling, and writing, impacts up to 20% of the global population. Traditional research methods, such as behavioral studies and neuroimaging, have provided valuable insights but face limitations in testing the underlying neurological mechanisms of reading impairments. To overcome these constraints, scientists from EPFL’s NeuroAI Lab employed next-generation Vision Language Models—AI systems capable of processing both visual and linguistic information—to simulate the full pipeline of word perception, from visual input to comprehension. In their approach, researchers first identified regions in the AI model analogous to the human visual word form area, which responds strongly to written words. They then selectively disrupted activity in these areas to observe the model’s behavior. The results showed that while the AI struggled with reading tasks, it retained the ability to understand images and language in broader contexts—mirroring the dissociation seen in human dyslexia, where reading difficulties coexist with intact verbal comprehension. This AI-driven model enables researchers to test hypotheses about dyslexia in a controlled, reproducible environment. One promising application involves evaluating how different font designs affect readability. By simulating how the AI brain processes text presented in various typefaces, scientists can pinpoint which fonts reduce cognitive load and improve word recognition—potentially guiding the development of more accessible digital and print materials. The study was presented at the 2026 International Conference on Learning Representations, a leading forum for advancements in machine learning and AI. Experts note that such models not only deepen scientific understanding of dyslexia but also pave the way for personalized educational tools. Future applications may include adaptive learning platforms that adjust text presentation in real time based on individual processing patterns. As AI continues to intersect with neuroscience, innovations like this offer hope for more effective, evidence-based support strategies for people with dyslexia—transforming how educators, clinicians, and technologists approach reading accessibility.
45