Autistic and non-autistic people seem to learn about others’ likes and dislikes in very similar ways, but autistic people’s tastes are so individual and varied that they’re harder for others to predict – and that, rather than a basic “social deficit,” may drive many everyday misunderstandings.
Rethinking social “misreads”
For years, autism has often been described through a deficit lens: autistic people supposedly struggle to read social cues, infer others’ thoughts, or understand what people want. A
new study in Nature Mental Health challenges that view by asking a simple question: how do we learn what someone else likes? The researchers argue that social friction may stem less from flawed learning and more from the sheer diversity of autistic people’s preferences, which makes “typical” social expectations a poor guide to any given autistic person.
Preferences as social building blocks
Being able to guess what someone likes – sushi or burgers, video games or hiking – is a basic ingredient of social life. It shapes how we choose gifts, plan outings, and decide whether we feel a connection with someone. The research team, led by psychologist Shannon Cahalan, used this “preference learning” as a precise way to study social cognition in autistic and non-autistic people, tracking how guesses change with feedback.
What people actually like
First, the team mapped out real preferences: 228 non-autistic adults, 125 non-autistic adolescents, and 255 autistic adolescents rated 120 foods and activities using a six-point emoji scale. Autistic teens showed a much wider spread of likes and dislikes than the other groups. On average, they rated candy, writing supplies, and art materials more highly, and gave lower ratings to salads, vegetables, and fitness equipment than non-autistic teens and adults.
Rigidity and “all-or-nothing” tastes
Researchers also looked at how consistently autistic teens organised their preferences. Many showed behavioural rigidity – a strong pull toward sameness – in how they rated items. If an autistic teen really liked one type of fast food, they often rated most fast foods similarly, forming tight clusters of likes and dislikes across related items.
Adults try to guess teen preferences
Next, 191 non-autistic adults were asked to learn about teenagers’ preferences – some autistic, some not – without being told who was who. Ninety-eight adults learned about non-autistic teens, and ninety-three about autistic teens. On each trial, adults guessed how much a teen liked an item, saw the true rating, and updated their mental model, allowing researchers to measure “prediction errors”.
Better at predicting non-autistic teens
Adults were more accurate when predicting non-autistic teens’ preferences, producing smaller errors overall. When they guessed about autistic teens, their initial errors were larger, though accuracy improved steadily with feedback. Computational modelling showed that adults used the same fine-grained learning strategy in both cases, updating beliefs based on similarities between items rather than relying on crude categories.
Autistic teens as social learners
In the final experiment, autistic adolescents did the guessing. Eighty-three autistic teens predicted the preferences of non-autistic teens, and 119 autistic teens predicted the preferences of other autistic teens, again without knowing who was autistic. The expectation was that autistic teens would do better with autistic peers, assuming shared experiences would make those preferences easier to read.
A surprising twist
Instead, autistic adolescents were more accurate when predicting the preferences of non-autistic teens than of other autistic teens. Just like adults, they used feedback to refine their guesses and relied on the same nuanced learning strategy. This challenges older claims that autistic people mainly project their own preferences onto others rather than building a distinct model of another person’s mind.
The key insight: variability, not deficit
Lead author Shannon Cahalan highlights “variability” as central to interpreting these findings. Rather than revealing a distinct, uniform “autistic” learning style, the study shows how different autistic people are from one another. Because autistic preferences are so diverse, an “average autistic profile” becomes almost useless when trying to predict what a specific autistic teenager likes. Both non-autistic adults and autistic teens found non-autistic preference patterns easier to learn, simply because they were more standardised and less scattered.
Traits that shape learning
Within the autistic group, certain traits influenced learning performance. Autistic participants with higher overall levels of autistic traits tended to update their beliefs more slowly during the task. Those with higher behavioural rigidity made more errors, suggesting that a strong preference for sameness can make it harder to flexibly revise expectations about others.
Limits and future directions
The authors note several limitations. Autistic participants were adolescents, whereas the non-autistic comparison group in the learning task was made up of young adults, so age differences may have influenced results. A small number of autistic teens could not complete the demanding task, meaning the study may underrepresent those with higher support needs. There were also relatively few autistic girls, making it hard to draw strong conclusions about gender.
What this means for everyday life
Outside the lab, the message is simple but powerful: autistic and non-autistic people seem to use very similar learning “rules” to understand others’ likes and dislikes. Misunderstandings arise because autistic preferences are more individual, making it harder for anyone – autistic or not – to rely on quick social shortcuts. Instead of seeing these misreads as proof of a social deficit, the study suggests we should recognise that our social norms are tuned to a relatively narrow idea of what’s “typical”. In autism, variability is a feature, not a bug – it complicates prediction, but reflects the rich diversity of autistic lives and interests.
