Summary
For thirty years, the dominant explanation for autistic social difficulties was simple: autistic people lack Theory of Mind — the ability to understand what others think and feel. The problem was located entirely in the autistic person. The solution was social skills training to make them more like neurotypical people.
In 2012, Damian Milton proposed a different explanation: the difficulty is bidirectional. Autistic people struggle to read neurotypical people; neurotypical people struggle to read autistic people. Empathy breaks down across a gap in neurotype, in both directions.
This reframing has become one of the most important conceptual shifts in autism research because it changes what needs fixing. A one-sided deficit model leads to teaching autistic people to perform neurotypicality. A bidirectional mismatch model requires changing neurotypical people’s behaviour and environments, not just autistic people’s.
What the evidence shows
The diffusion chain studies
The most striking empirical support comes from Catherine Crompton’s diffusion chain experiments (2020, Autism; 2025, Nature Human Behaviour).
In the original 2020 study, groups of eight passed a story through a chain like telephone. Pure autistic chains and pure neurotypical chains retained information equally well. Mixed chains (alternating neurotypes) degraded information significantly.
Rapport told the same story. Same-neurotype chains reported higher rapport than mixed chains. Autistic people communicating with autistic people felt as connected as neurotypical people communicating with neurotypical people. The breakdown was at the interface.
The 2025 registered report (311 participants, three universities, published in Nature Human Behaviour) largely confirmed these findings and added a nuance: rapport was higher with same neurotype and when autism diagnosis was disclosed — suggesting that knowledge of neurotype difference can partially bridge the gap.
Thin-slice judgements
Sasson and Morrison (2017, Scientific Reports) documented the other half of the problem. Neurotypical participants formed negative first impressions of autistic people from brief video clips — within seconds. They rated autistic individuals as less likeable, less attractive, and less desirable as social partners.
When the same conversations were presented as text (no audio-visual cues), the biases disappeared. Neurotypical rejection targeted autistic style—prosody, facial expression, body language—not substance. The content was fine; the delivery triggered rejection.
This is the double empathy problem in action: neurotypical social cognition misreads autistic social signals, just as autistic social cognition misreads neurotypical ones. Both sides are doing their best with a perceptual system calibrated for their own neurotype.
The empathy meta-analysis
Cheang et al. (2025, Autism) conducted a systematic review of 226 studies on empathy in autism. The findings challenge the “autistic empathy deficit” narrative:
Cognitive empathy (understanding what someone else thinks) showed a large effect size difference (g = −0.85). Affective empathy (feeling what someone else feels) showed only a small effect (g = −0.17). Autistic people feel for others nearly as much as anyone; they process the thinking about others’ states differently.
The distinction matters enormously. The “mindblindness” framing implied autistic people don’t care about others. The evidence says they care but process social information through different channels.
The sensory dimension
The double empathy problem is not just about “understanding minds.” It is about inhabiting different sensory umwelten in the same social space.
In a café conversation: a neurotypical person’s auditory system filters background noise. An autistic person’s processes it at nearly equal volume to speech. The neurotypical person reads subtle facial cues signalling “I’m about to speak.” The autistic person’s visual processing tracks a different set of details. The neurotypical person doesn’t notice the fluorescent light; the autistic person’s processing capacity is consumed by it.
Same room, same conversation, different perceptual worlds. The breakdown is not about empathy. It is about two umwelten failing to align, with neither person realising how different the other’s sensory reality is.
Open questions
Does the double empathy problem apply equally when intellectual disability is present? Milton’s framework assumes both parties have broadly comparable communicative capacity. For autistic people with significant ID, the picture is more complicated. The mismatch is real, but so are genuine communicative limitations that cannot be resolved simply by meeting in the middle.
Can neurotypical people be trained to read autistic social signals, the way autistic people are currently trained (through social skills programmes) to read neurotypical ones? The evidence suggests this would be at least as effective — and more equitable.
How does the double empathy problem interact with masking? If an autistic person is performing neurotypical social behaviour — see Masking and camouflaging — the neurotypical person may believe communication is succeeding when the autistic person is actually experiencing it as exhausting and inauthentic.
Implications for practice
If you design social support for autistic people: the goal should not be “teach them to be more neurotypical.” It should be “create conditions where communication succeeds across neurotypes.” This means: smaller groups, quieter environments, explicit communication norms, and — where possible — connecting autistic people with other autistic people.
If you are neurotypical and interact with autistic people: the communication difficulty you experience is symmetrical. You are as hard for them to read as they are for you. Adjusting your own communication style (being more explicit, reducing reliance on facial cues, allowing processing time) is as important as expecting them to adjust theirs.
If you are autistic: the research confirms what many autistic people already know — communication with other autistic people often feels easier, more natural, and more satisfying. This is not a consolation prize. It is evidence that your social cognition works well in compatible contexts.
Key sources
- Milton, D. (2012). “On the ontological status of autism: the ‘double empathy problem.‘” Disability & Society, 27(6).
- Milton, D., Gurbuz, E., & López, B. (2022). “The ‘double empathy problem’: Ten years on.” Autism, 27(1).
- Crompton, C.J. et al. (2020). “Autistic peer-to-peer information transfer is highly effective.” Autism, 24(7).
- Crompton, C.J. et al. (2025). “Information transfer within and between autistic and non-autistic people.” Nature Human Behaviour, 9.
- Sasson, N.J. & Morrison, K.E. (2017). “Neurotypical Peers are Less Willing to Interact with Those with Autism.” Scientific Reports, 7.
- Cheang, R.T.S. et al. (2025). “Do you feel me? Autism, empathic accuracy and the double empathy problem.” Autism, 29(5).