Whether the difference lives in building the map or in the choices of navigating it
Sarah Banker, Daniela Schiller, Jennifer Foss-Feig and colleagues at Mount Sinai published a study in Biological Psychiatry in 2026 that set out to locate, in the brain, the social difficulty associated with autism. They used a validated “social navigation” task: participants move through a choose-your-own-adventure narrative, interacting with six characters, and each choice quietly shifts those characters’ positions within a two-dimensional social space defined by affiliation (warmth) and power (hierarchy). Using 7-tesla fMRI, the high end of spatial resolution, the researchers can measure both where a participant places every relationship and which brain regions are doing the work while they do it. They recruited 122 adults across three groups — 42 autistic, 46 typically developing, and 34 with misophonia as a clinical comparison — the last group letting them separate what is specific to autism from what comes with social difficulty in general.
The design’s real value is that it pits two very different accounts of autism against each other. The first holds that autistic adults fail to build accurate social maps at all — that the underlying representation of who-stands-where is itself impaired. The second holds that they build the maps fine but differ in how they apply them at the moment of a social decision. One account locates the difference deep in the representational machinery; the other locates it at the point of action, leaving the map intact.
These are not equivalent claims dressed differently. They are rival ontologies of what autism is at the neural level — a difference in the seeing, or a difference in the acting. And they are not equally flattering to the deficit model, because one says the autistic social brain constructs a defective world-model while the other says it constructs the same model and moves through it differently. The task was built to tell them apart, by measuring both the neural tracking of social distance as it updates trial by trial and the neural activity at the moment of each choice. The data came down clearly on one side.
The representational machinery returns indistinguishable across every group
The map-building result is the headline, and it is a null in the most informative sense. When the researchers looked at neural tracking of social distance — the brain dynamically registering how close or far each character sits, updating as the interaction unfolds — they found no group differences at all. Autistic adults tracked social distance in the same regions, to the same degree, as everyone else: the posterior cingulate cortex, the precuneus, the parahippocampal place area, the same network that does this work in non-clinical samples. The authors state plainly that they found insufficient evidence to support impaired social map creation in autism.
What differed was narrower, and needs reporting precisely rather than inflated. First, a behavioural difference: autistic adults, on average, placed characters at greater social distance, driven specifically by lower affiliation — choices that kept characters further away. Second, a neural difference confined to the moment of decision: reduced activity in the left temporal pole, a region involved in retrieving and applying social knowledge, which the authors link to smaller real-world social networks. The machinery that builds the map is shared and intact; something at the point of consulting it differs.
Here the paper’s framing and my own part company, and I would rather flag that than blur it. The study reads the greater social distance as impairment — as “aloofness,” a failure of affiliation. That reading is available, but the task cannot distinguish it from a very different one: that autistic people, navigating the narrative, simply made different choices about how close to let characters come — preference, or calibration to a lifetime of interactions that have gone less warmly, or an accurate reading of the specific characters. The behaviour is “more distance.” The interpretation “less capable of closeness” is added on top, and the data does not force it. “Applies social knowledge differently” and “is impaired at being social” are not the same claim, and only the first is what the scan actually shows — the second is a value laid over the measurement, and it’s important to see that as a choice rather than a finding.
What the scan cannot see, and the reading it reaches for anyway
There is one further thing worth naming precisely because the paper does not. The behavioural result — autistic adults keeping more distance, being less affiliative — is exactly the kind of finding the double empathy problem was formulated to reinterpret: the breakdown in autistic social interaction as relational and bidirectional, a mismatch between two parties, rather than a deficit sited inside one of them. A paper finding that autistic people choose more social distance is standing on the doorstep of that literature, and it does not walk through the door.
It is now that I must be honest about why it structurally cannot. This is a single-brain study: one autistic person, alone in a scanner, navigating a scripted narrative with characters who are not real and cannot respond. The double empathy claim is about what happens between two minds, and a between-minds claim cannot be tested by imaging one mind interacting with a script. So the paper’s silence is not quite an oversight — the design has no purchase on the question.
But the silence is still telling, because the reading the paper does reach for — “aloofness” — is a single-brain, deficit-sited interpretation of a behaviour the relational literature would read entirely differently, and the paper adopts it without registering that the alternative exists. The greater social distance gets explained as something wrong with the autistic participant, when the task was never built to locate the difficulty in a person versus a relationship at all. None of this diminishes what the study genuinely establishes, and it is worth ending on that: within a high-IQ adult sample — the authors are clear the finding may not extend to autistic people with higher support needs — the machinery for building social maps came back intact, indistinguishable from controls, and the difference narrowed to behaviour and a single decision-stage region. Read straight, that is more anti-deficit than the paper’s own framing allows. The social brain, at the level of representation, is not broken. What autistic people do with the map differs — and whether “differs” means “impaired” is the question the data leaves open, answered by the paper in one direction without noticing it had a choice.
Citations
Banker, S. M., Schafer, M., Barkley, S. et al. (2026) — Neural Tracking of Social Navigation in Autism Spectrum Disorder — Biological Psychiatry
Tavares, R. M., Mendelsohn, A., Grossman, Y. et al. (2015) — A Map for Social Navigation in the Human Brain — Neuron
Milton, D. E. M. (2012) — On the ontological status of autism: the ‘double empathy problem’ — Disability & Society
Schafer, M. & Schiller, D. (2018) — Navigating Social Space — Neuron
