Pinpointing theory of mind deficiencies in autism
by Amy McDermott
Around age four, preschoolers learn that other people have unique thoughts and feelings, an inner life. These children start to be able to predict a person’s behavior based on that understanding. They start developing theory of mind (ToM).
That developmental process doesn’t always go smoothly — ToM deficiencies are often, for example, associated with autism spectrum disorders. Exactly how the multifaceted phenomenon of ToM breaks down is a mystery, however. Now researchers, in a recent study in Current Biology, have taken steps to distinguish between the facets of theory of mind, to better understand how and where these neurological processes might be impaired in autistic adults.
Psychologists historically bundled theory of mind into one cognitive process, explains Liane Young, a social psychologist and neuroscientist at Boston College. “Now folks are converging on the notion that theory of mind can be decomposed into multiple component parts,” she says, “and we’re trying to tease apart how those parts might work together.”
The new study approaches ToM as a complex developmental process with multiple aspects: imagining what another person knows, predicting what they want, and understanding their actions in the context of their belief. Autistic adults, it found, struggle to use another person’s beliefs to understand their actions, by intuiting their intentions. Dissecting the learning process offers a more nuanced understanding of the exact problem, says neuroscientist Damian Stanley of Adelphi University in Garden City, NY, who coauthored the study (which was conducted at Caltech in Pasadena, CA).
First, as part of a simple computer game, 26 high-functioning adults on the autism spectrum, and 53 adults without autism, made decisions about whether or not to donate to a charity. In each round of the game, they could choose to donate to a charity (one of three options appeared per round), or keep the money for themselves. But there was a twist: sometimes the mode of the computer game reversed for several rounds without warning. In reverse mode, if a subject clicked the button to donate to a charity, for example, most of the time the opposite would occur — they kept the money for themselves.
In a second task, the participants saw a woman’s face on one side of the screen, and one of the three charities on the other side. The subjects knew that the woman was playing the charity game, with the same rules and knowledge they’d had, and the same unpredictable mode changes. They had to guess if the woman donated to charity, as well as which mode she thought the game was in. Importantly, the subjects themselves knew the mode, indicated by a circular icon at the top of their screen.
In the first few rounds, all of the subjects were bad at guessing the woman’s intentions with the money. But over time, the non-autistic subjects improved their predictions by tracking which charities the woman preferred, if any, and how her choices changed with her beliefs about the game’s mode. And both autistic and control groups were pretty good at guessing which mode the woman thought the game was in.
But the autistic participants struggled to integrate the woman’s beliefs with her past choices, and therefore failed to learn her intentions for the money.
Following the experiment, the researchers fit the data to computational models, to figure out if the participants used particular strategies to make predictions. The control group did: People without autism tracked the woman’s beliefs and choices over time to figure out her intentions. They assumed her intentions stayed the same over time, and used them to make predictions of her future behavior, Stanley says.
Participants with autism, however, didn’t do that, as they struggled to use the woman’s beliefs to learn her intentions. But no single model explained their behavior. Three models proved promising for autistic subjects, Stanley says, meaning there were multiple ways they could struggle. Autism is a heterogeneous disorder, so the variation wasn’t too surprising. “We wouldn’t expect there to be one specific way in which individuals with autism spectrum disorders are impaired,” he says, “but it provides us with a more nuanced target for further examination.”
This is among the first studies to use computational modeling to examine autism, and the first to tease apart belief and intent to track the evolution of decision-making in a modeling context, the researchers say. Young applauds the study for not just “looking at static theory of mind, but looking at learning over time.”
Future work with more subjects, and computer tasks designed specifically to investigate the three most promising behavioral models, may “tease out subgroups of people on the autism spectrum who use different specific learning strategies to predict other people’s behavior or preferences,” Stanley says. Imaging could also reveal the brain regions where these mental processes take place in autistic versus non-autistic subjects.
Originally published at blog.pnas.org on February 8, 2019.
