A team of researchers at Stanford University combined the power of artificial intelligence and functional magnetic resonance imaging (fMRI) to create “digital twins” of struggling mathematics students to provide first-time insights into the neurological foundations of mathematics learning disabilities.
Vinod Menon, professor of psychiatry and behavioral science at Stanford University, said: “Now we have these new AI tools and we can actually ask these questions deeper and far more mechanically.”
In a paper published in the journal Advances in scienceMenon and his co-authors, Dr. Stanford Postdoctoral Dr. Anthony Strock and social science researcher Percy Mistry, introduce what is called a personalized deep neural network. These are effective digital “two-brains” for real children, models that mimic how individual students solve mathematical problems and can calculate where things get wrong in the brain of a child with a mathematics learning disorder.
In din
In their study, funded in part by Stanford University, human-centered AI, researchers selected 45 students between the ages of 7 and 9, of which they had mathematics learning disabilities. The actual students then solved basic addition and subtraction problems, while fMRI charted brain activity. Next, the AI model simulates brain activity, but gets answers similar to actual twins, similar to actual twins, and is correct and wrong every time.
Menon and his collaborators have learned that AI models can be tailored to mimic the accuracy and learning speed of real-world twins. This is adjusted by adjusting a single neurological parameter known as neuroexcitivity, which is roughly equivalent to how the brain cells fire strongly. Such neurological understanding is difficult to study in living subjects and requires electrodes placed in the brain to measure neural activity. Therefore, the true neurophysiology of learning disabilities and learning disabilities was elusive to identify scientifically.
“Contrary to what we and others were expecting, we found that not too much neural activity is a core problem for learning difficulties,” Menon said. “Struggling children showed signs of hyperthermia in brain regions, the key to numerical thinking, and AI twins showed the exact same pattern.”
Menon and team's hypothesis is that this excess activity leads to confusing the mental representation of mathematical problems, confusion, and slow learning. They theorize that hyperexcitation leads to what they call representational overlap. Mathematical problems produce neural patterns that are too similar. Mathematical representations are mixed and confused, Menon said, preventing accurate problem solving. It's as if the brain is screaming for itself, and students cannot identify the correct answer in the fuss.
New hope
The educational impact is substantial. Digital twins allow researchers to test neurological mechanisms In Silico – With a computer – Provide each child with a window into the brain-level causes of learning struggle. Menon highlighted that in this study, the AI Twins modelling for mathematics learning disabilities shows that nearly twice as much training is required to reach the same accuracy as regular mathematics students. But Menon said, “They ultimately reach comparable performance, and that gives us great hope for an improved repair strategy.”
For educators, digital twins can lead to personalized learning plans tailored to a particular student's learning style, and predict the best type of instruction for each learner. Menon and the team are now expanding their models in new directions to create even neurological simulations of mathematical inference.
The point of Menon is that children with learning disabilities may need important additional training that can help improve performance shortages. Nevertheless, Menon is careful not to overload the results. Models need refinement. There is more to do, but it points to some promising new directions for further research.
“We have a framework to test target strategies before trying them out in real classrooms. “It can accelerate our ability to design effective educational programs for children with learning disabilities and make real progress for real children struggling to learn mathematics.”
Vinod Menon is a professor of psychiatry and behavioral sciences in neurology, neuroscience and education at Stanford University, and professors of psychiatry and behavioral sciences, Rachel L and Walter F. Nichols, Maryland He is also the director of the Stanford Institute for Neurosciences in Cognition and Systems, and an affiliate for the Wu Tsai Neurosciences Institute and the Stanford Institute for Human-Centered AI. Research authors include Stanford postdoctoral scholar Anthony Strok and Stanford University research scholar Percy Mistry.
