Diagnosing Autism More Precisely

But the people working inside Colby’s new Davis Institute for Artificial Intelligence are focusing more on opportunities to use AI to improve lives. When the institute opened two years ago, thanks to a $30-million gift from the Davis family and trustee of its charitable foundation, Andrew Davis ’85, LL.D.’15, it became the first cross-disciplinary institute for AI at a liberal arts college.

“One of the beauties of people who are trained in the liberal arts is that they really understand how to come at a problem from multiple, different angles,” President David A. Greene told National Public Radio after the institute opened. “I think the more that we have people who are coming from liberal arts backgrounds, who are really raising the kind of questions that will ultimately shape AI in more positive ways, the better off we’ll be.”

The Davis Institute for Artificial Intelligence has become the home of intense interdisciplinary collaboration. Approximately 25 percent of faculty representing all four academic divisions teach AI courses that reach close to 20 percent of the student body, and the institute hosts scholars, speakers, and events.

When psychologists attempt to determine whether a child is autistic, Chowdhury said, “There is no physical test, there is no chemical test you can perform. All you look for is behavioral markers.”

AI, Chowdhury said, can help psychologists identify more of those behavioral markers and make more accurate diagnoses. And incorporating AI into the diagnostic process will allow psychologists to focus more intently on their young patients.

“The way autism is now diagnosed is psychologists give the children or the individuals being tested a couple of questions or tasks,” she said. “They need to look at how the individuals respond to these tasks. So that means they have to look at what they’re saying, how they’re saying it, and how they’re expressing their emotions. The psychologists are making notes at the same time. It’s cognitively demanding. My hope is, through the type of work that I’m interested in, we will help augment the practitioner’s work.”

In short, practitioners could rely on AI to catch what they might miss in that scenario. Davis AI Director Amanda Stent elaborated.

“The psychologist who does the conversation has to do four things at once, which is three too many,” Stent said. “They have to conduct the conversation in the structure that is given to them by the diagnostic manual. They have to take notes on the conversation. They have to monitor themselves and their own behaviors. And they have to monitor the child’s behavior and attempt to make a diagnosis.”

Chowdhury and Romero are working together to build an AI tool that will automatically analyze the video recordings of the conversations. The idea is that an AI tool could do the note-taking and record the behaviors, allowing the psychologist to focus on the conversation.

Romero has worked with children and adolescents with autism spectrum disorder since 2013. Her dissertation research investigating the “Computational Measurement of Social Communication Dynamics in Adolescents with Autism Spectrum Disorder” was funded by a predoctoral fellowship from the National Institutes of Health.

For years, she has focused on improving the demanding diagnostic process by finding ways to automate certain aspects to lighten the cognitive load of clinicians. Romero began thinking about incorporating AI tools while working at the Davis Institute for AI as one of its first sabbatical fellows.

The collaboration with Chowdhury has led to publications and a fall workshop, Connecting Multiple Disciplines to AI Techniques in Interaction-centric Autism Research and Diagnosis.

Romero is reaching out to the autism community in Maine and recruiting to further these efforts in her research. If you or someone you care about has autism and would like to discuss Romero’s research, please contact her at veronica.romero@colby.edu.