Many children struggle with learning difficulties and while they may be hurdles to be overcome with support from parents and schools, this is not an impossible task. A new study suggests that the current broad categories struggling children are placed in often fails to provide enough details to properly help these kids, and they've found a new way to more accurately identify learning difficulties using artificial intelligence (AI).
Researchers at the University of Cambridge’s Medical Research Council (MRC) Cognition and Brain Sciences Unit used a state-of-the-art AI approach to analyze data from 550 children, mostly between seven and 12 years, who had been referred to the Centre for Attention Learning and Memory.
The researchers pointed out that most studies on children's learning difficulties were based on children who had already been given a diagnosis – ADHD, autism spectrum disorder, dyslexia etc. But their machine learning algorithm showed that this is limited. The diagnostic label is too broad to give enough information to actually support the kids as it doesn’t come close to pinpointing the reason why they are struggling. One child’s ADHD is not like another’s child’s ADHD, they state as an example.
"The machine learning shows that there are different routes to being a struggling learner. Having a diagnosis – whilst important for children and families – does not inform practitioners about which particular route a child has taken. But knowing this is vital if they are to receive proper tailored support,” lead author Dr Duncan Astle told IFLScience.
The study, published in Developmental Science, is the first one to use a machine-learning algorithm approach on a broad spectrum of hundreds of struggling learners, including ones who had not been previously diagnosed with a specific disorder. They fed the AI data on each child including on vocabulary, listening, spatial reasoning, problem-solving, and memory. From this, it clustered the children into four categories of difficulties. This matched other data on the children, such as school reports on reading and math skills, as well as parental reports on communication ability, but it didn’t appear to correspond with their diagnoses.
The four clusters are children with broad cognitive difficulties, children struggling with processing sounds in words, children with difficulties with working memory (short-term retention of information), and children with typical cognitive results for their age. The latter group suggests how behavioral difficulties, not included in the machine learning analysis, play a role as well.
“Our work suggests that children who are finding the same subjects difficult could be struggling for very different reasons, which has important implications for selecting appropriate interventions,” senior author Dr Joni Holmes explained in a statement.
The team investigated if the clusters division came from underlying biological differences by performing MRI scans on 184 of the children. The grouping matched some patterns in connectivity in the children’s brains, a hint that there could be a biological cause.
“These are interesting, early-stage findings which begin to investigate how we can apply new technologies, such as machine learning, to better understand brain function,” Dr Joanna Latimer, Head of Neurosciences and Mental Health at the MRC, concluded.