Gabriela Rosenblau and her team at George Washington University have upended a common assumption about how autistic and non-autistic people understand each other. Their new research, published in Nature Mental Health in June 2026, reveals that when it comes to learning about one another's preferences, both groups actually rely on the same fundamental strategies — but a crucial difference lies not in how they learn, but in what they're trying to predict.
Understanding how different groups of people make sense of each other matters deeply. For years, researchers and the broader public have questioned whether communication breakdowns between autistic and non-autistic individuals stem from fundamentally different ways of thinking. If true, it might suggest that bridging these gaps requires completely different approaches. Rosenblau's work suggests something more nuanced and, paradoxically, more hopeful: the problem isn't that autistic and non-autistic people learn differently — it's that autistic individuals simply have more varied preferences, making them harder to predict using standard social assumptions.
The study, led by Rosenblau, an associate professor of cognitive neuroscience and director of the Cognitive Neuroscience Doctoral Program, along with Ph.D. student Shannon Cahalan, involved recruiting large samples of both autistic and non-autistic participants for an online experiment. Researchers first captured each person's individual preferences for foods and activities. Then they tested how well non-autistic adults and autistic adolescents could predict the preferences of both groups.
The findings were striking. Autistic participants showed significantly greater variability in their personal preferences than their non-autistic peers — a reminder that autism itself is remarkably diverse, and trying to predict what any individual autistic person will prefer based on group patterns is inherently unreliable. Both autistic and non-autistic participants performed worse when trying to predict autistic adolescents' preferences than non-autistic adolescents' preferences. Yet when researchers examined the underlying learning mechanisms, they discovered something unexpected: both groups used nearly identical strategies to make their predictions, and both groups updated their guesses when given feedback.
One wrinkle emerged: as autism severity increased, participants showed lower learning rates and struggled more to update their predictions based on new information. But the core insight remained stable across the data: the learning strategies themselves were fundamentally similar.
"These results suggest that misunderstandings between autistic and non-autistic people may not stem from fundamentally different learning mechanisms," Rosenblau said. "Instead, they may arise because autistic individuals' preferences are more varied, making them harder to predict using typical social assumptions."
This finding offers new support for the "double empathy problem," a theory suggesting that communication barriers between autistic and non-autistic people don't arise from a lack of empathy in autistic individuals, but rather from differences in how each group interprets the social world. If both groups learn similarly, the real work lies not in teaching one group to think like the other, but in recognizing that autistic individuals' richer diversity of preferences means assumptions won't work — and that's not a deficit, it's simply reality.
The research also points toward a path forward: Rosenblau emphasized the value of combining large datasets with computational modeling to understand social learning in autism more deeply. This framework may help identify meaningful differences within the autism spectrum itself, potentially informing more tailored and effective interventions in the future.