Study Reveals Preference-Learning Similarities and Variability in Autism

A three-experiment study published in *Nature Mental Health* examined how autistic and non-autistic individuals infer others’ likes and dislikes. Participants rated 120 food and activity items, then guessed peers’ ratings while receiving immediate feedback. Across 228 non-autistic adults, 125 non-autistic adolescents, 255 autistic adolescents, 83 autistic teens guessing non-autistic peers, and 119 autistic teens guessing autistic peers, researchers found that both adults and autistic adolescents employed a fine-grained learning strategy that updates guesses based on item similarity. Adults produced smaller prediction errors when guessing non-autistic teens, but improved over time with autistic teens. Autistic adolescents, contrary to expectations, were more accurate predicting non-autistic teens than autistic peers.

Psypost

The work builds on the “double empathy problem,” which posits that communication breakdowns arise because autistic and non-autistic people experience the world differently, rather than reflecting a deficit in autistic individuals. Earlier research by Dr. Gabriela Rosenblau (2021) using preference-learning paradigms suggested distinct inference strategies between groups; the current study expands that inquiry with larger, more diverse samples.

Psypost

• Preference mapping: autistic adolescents displayed the widest variety of likes, rating candy, writing supplies, and art supplies higher, and salads, vegetables, and fitness equipment lower than non-autistic groups.

Psypost

• Experiment 2 (191 non-autistic adults): 98 guessed non-autistic teens, 93 guessed autistic teens; prediction errors were initially larger for autistic teens but decreased with feedback.

Psypost

• Experiment 3 (202 autistic teens): 83 guessed non-autistic teens, 119 guessed autistic teens; accuracy was higher for non-autistic targets.

Psypost

• Higher self-reported autistic traits correlated with slower integration of feedback; greater behavioral rigidity predicted more overall errors.

Psypost

The findings suggest that social misunderstandings stem less from an inability to read cues and more from the high variability of autistic preferences. Because “average” autistic preference profiles fail to capture individual nuances, both neurotypical and autistic observers find autistic peers harder to predict. Recognizing variability as a “feature” rather than a “bug” may shift interventions toward personalized, rather than stereotyped, social learning approaches.

Psypost

Lead author Shannon Cahalan, a postdoctoral fellow in the National Institute of Mental Health’s Section for Social and Developmental Cognitive Neuroscience, emphasized that the study challenges the notion that autistic youth rely solely on their own likes when judging others. She noted that the research highlights the importance of individual differences and plans to extend the work to neural imaging to explore functional activity underlying preference learning. The study’s co-authors—Raphael Perla, Sophia Block, Mikaila Loughlin, Christoph W. Korn, and Dr. Gabriela Rosenblau—stress that the extensive behavioral dataset will support future analyses of how demographic and neuropsychological factors shape social learning.

Psypost

• Age mismatch: comparing adult learners with teenage targets confounds developmental effects with diagnostic status.

Psypost

• Task complexity: a subset of autistic participants could not complete the guessing game, limiting generalizability across the spectrum.

Psypost

• Gender imbalance: only 55 autistic girls participated, reducing power to detect sex-based differences.

Psypost

• Lack of diagnostic disclosure: participants were unaware of peers’ diagnostic status, which may have altered learning strategies.

Psypost

• Restricted rating scale (six emoji-based options) limited the granularity of prediction-error calculations.

Psypost

No direct contradictions appear among the presented data, but the study leaves unanswered questions about how explicit knowledge of diagnostic status, broader age ranges, and more nuanced rating systems would affect learning outcomes.

Psypost

• “The senior author, Dr. Gabriela Rosenblau, published findings in 2021, using the preference learning paradigm and computational modeling, which found that autistic and non-autistic participants adopted different strategies when inferring peer preferences,” — Shannon Cahalan, Postdoctoral Fellow, NIMH

Psypost

• “Specifically, non-autistic youth’s learning could be described by a complex model integrating prior knowledge about peers’ preferences, item relationship representations, and trial-by-trial updating while autistic youth’s own preferences informed their inferences about peers,” — Shannon Cahalan

Psypost

• “I was quite surprised as to how integral ‘variability’ was to the interpretation of our results,” — Shannon Cahalan

Psypost

• “Variability is a ‘feature’ of autism and not a ‘bug,'” — Shannon Cahalan

Psypost

• “On the other hand, variability in autistic traits, like rigidity for example, were especially predictive of autistic teen’s preference learning performance.” — Shannon Cahalan

Psypost