A between-minds question the scanner structurally could not reach
I recently covered a 7T neuroimaging study which found that autistic adults build social maps exactly like everyone else — and I noted that a single-brain design structurally cannot test the double empathy problem, because Milton’s claim is about what happens between two minds, not inside one. Zachary Miller, Caitlin Conner, Judy Chang and Carla Mazefsky, at the University of Pittsburgh and Michigan State, have now published the complement in the journal Autism: instead of imaging one mind interacting with a script, they asked autistic adults directly what conversation costs across partner neurotypes.
The instrument is the Neurotype Communication and Assistive Technology Survey, designed for the study and refined with feedback from four autistic adults. They surveyed 165 autistic adults aged 18 to 35, analysing 154 responses, mixing Likert ratings with open-ended descriptions across four domains — environment, familiarity, partner neurotype, and emotional state — with each domain assessed for both matched and mixed-neurotype interactions.
The design is the experiential inverse of the scanner study, and the pairing matters. Neuroimaging can show that the machinery is intact but cannot ask what an interaction feels like; a reflective survey can capture exactly that but cannot verify mechanism. Neither alone settles the double empathy question. But together they triangulate it — the brain data showing no representational deficit, the lived data showing where the difficulty actually accumulates: in the crossing between neurotypes, not in the autistic mind alone.
What conversation costs depends on who is on the other side of it
The neurotype findings run exactly as the double empathy problem predicts. Participants reported that conversations with allistic partners required more conscious effort, more social masking, and more cognitive load than conversations with autistic partners. In the qualitative responses — provided by 85% of the sample — communication with autistic people was described as direct, specific and efficient, with the most frequent themes being mutual intersubjectivity and authentic self-expression. Communication with allistic friends, by contrast, was dominated by misunderstandings: vague wording, figurative language, sarcasm, tone, facial cues.
The participants’ own words carry the asymmetry better than the statistics. One described navigating allistic indirectness as requiring “mental gymnastics.” Another said communication with autistic family members is more direct and avoids ambiguity. And the authority-figure finding extends it into consequence: 62.3% reported difficulty communicating with supervisors and teachers, with one participant describing having to put on “even more of a show” with authority figures — thought processes consumed by how the performance is landing rather than by the conversation itself. That is masking described from the inside, at the exact site where employment and education outcomes get decided.
Familiarity compounded the effect rather than replacing it. Ease was highest with autistic family members and lowest with allistic friends and strangers — neurotype alignment and familiarity working jointly, which fits the prior finding that familiarity reduces the cognitive load of interaction by allowing reliance on established patterns.
Honesty requires the full picture, and two details complicate the clean story. First, participants self-identified as autistic — no clinical diagnosis was required — which the authors flag, and which adds heterogeneity to who exactly is being described. Second, the wider literature is not unanimous: the well-known finding that information transfers more efficiently between autistic people has not consistently reproduced in recent replication work, which the authors themselves cite. What this study adds is not proof of superior autistic-to-autistic transmission — it is systematic evidence of differential cost: whatever is happening to the information, the effort, masking and load are reliably higher on the mixed-neurotype crossing. The difficulty lives in the mismatch. That much survives the replication debate intact.
The room and the feeling change what is possible to say
The environmental findings relocate another piece of the difficulty out of the person. A majority — 56.4% — reported that surroundings at least somewhat affect their ability to communicate. In busy and public settings, the dominant reported experiences were going quiet and losing words (mentioned by over half of those describing such settings) alongside general verbal difficulty. In familiar, personal environments, the dominant theme flipped to natural, expressive, open communication. Work and school sat in between, with their own signature: anxiety, scripting, and masking. Same person, different room, different capacity — one participant describing how overwhelming stimuli make it harder to concentrate on words and properly convey meaning at all.
There’s a practical corollary in the modality data: participants whose communication suffered most from the environment leaned significantly more on text-based channels — email, messaging, AAC-related strategies — while telephone showed no association at all. The authors are careful that this doesn’t mean text-reliance causes difficulty; it means people whose speech is environmentally fragile maintain access through channels that don’t collapse under sensory load. For anyone designing workplaces or services, that is a finding about which doors to keep open.
Emotion ran through everything, and here the study produced its most honest result: a null. Negative emotional states worsened communication for 52.6% of participants — with loss of words or withdrawal reported in 94% of coded instances of feeling upset or overwhelmed — while positive states improved fluency and expressiveness, and calm states were associated with expressive communication in every coded instance. But contrary to the researchers’ own hypothesis, emotion’s impact did not differ significantly by partner neurotype. Distress impairs autistic communication with everyone, not just across the neurotype divide. The authors suggest cognitive overload as the mechanism — strong emotion depletes the resources conversation runs on, whoever is present. The double empathy gap is real, but it is not the whole account: some of the difficulty is the mismatch, and some is simply what an overloaded system can no longer afford.
What the data cannot settle, and what it shifts anyway
The limitations are the standard ones for this design, and worth stating plainly. This is retrospective self-report, vulnerable to recall difficulty — which the authors note may be amplified by alexithymia, itself more prevalent among autistic people. The sample is predominantly white, mostly low-income, aged 18 to 35, recruited through advocacy networks, and includes self-identified participants; online autism research also carries a documented fraud and bot risk the team mitigated but cannot eliminate. And the survey captures only one side of each remembered conversation — the allistic partners’ experiences, emotions and adaptations went unmeasured, which is an ironic limit for a study about mutuality.
But notice what the findings do even inside those limits. The standard account of autistic communication difficulty locates the problem in the autistic person: a pragmatic language deficit, to be remediated with social skills training aimed at making the autistic party perform allistic norms more convincingly. This data — cost concentrated at the neurotype crossing, fluency restored among autistic partners and in low-demand environments, distress impairing everyone equally — is not the shape a unilateral deficit would produce. A deficit doesn’t switch off when the listener changes.
The intervention implications follow directly, and the authors draw them: reduce sensory stress, support emotional wellbeing, and teach explicit cross-neurotype communication strategies — to allistic people as well as autistic ones. That last clause is the paradigm shift in miniature. If the difficulty lives in the crossing, then training only one side of the crossing is not support; it is the unilateral-deficit model wearing a lanyard. The double empathy problem, measured from the inside, says the work belongs to both parties — and that much of what looks like autistic communication impairment is the bill for conversations conducted on one neurotype’s terms, in rooms built for one neurotype’s senses.
Citations
Miller, Z. A., Conner, C. M., Chang, J. C. & Mazefsky, C. A. (2026) — Navigating Mixed-Neurotype Dialogue: Environmental and Affective Influences on Autistic Speakers — Autism
Milton, D. E. M. (2012) — On the ontological status of autism: the ‘double empathy problem’ — Disability & Society
Crompton, C. J., Ropar, D., Evans-Williams, C. V. M. et al. (2020) — Autistic peer-to-peer information transfer is highly effective — Autism
Crompton, C. J., Foster, S. J., Wilks, C. E. H. et al. (2025) — Information transfer within and between autistic and non-autistic people — Nature Human Behaviour
Heasman, B. & Gillespie, A. (2019) — Neurodivergent intersubjectivity: Distinctive features of how autistic people create shared understanding — Autism
