I posted a measured response to Litman et al.’s recent Nature Genetics paper, which has been widely circulated as showing autism is “at least four conditions.” 2 My point was straightforward: the study shows autistic heterogeneity is organized, not random, but we shouldn’t move from one oversimplification — autism is one thing — to another — autism is four things. The better shift is from autism-as-flat-diagnostic-label to autism-as-emergent-developmental-landscape: different trajectories shaped by genetics, biology, regulation, environment, and timing.
2. Litman, A. et al. (2025). Decomposition of phenotypic heterogeneity in autism reveals underlying genetic programs. Nature Genetics. DOI: 10.1038/s41588-025-02224-z. The paper itself identifies four phenotype classes within a heterogeneous distribution via general finite mixture modeling — a statistical move, not an ontological one. The “autism is four conditions” framing belongs to the paper’s reception, not to the paper.
A researcher I respect — someone with deep credibility in the field, and someone who is also autistic — commented on the post. The substance of her comment was that the principle I’d articulated isn’t autism-specific at all: how is this different for autistic and non-autistic people? It is not. WHEN HAS A ONE-SIZE-FITS-ALL model ever worked for anything?
The comment used emphatic punctuation and stretches of capital letters to drive the points home.
I read it and froze. Was it sarcasm? Pushback? I genuinely couldn’t tell.
The all-caps. The forceful rhetorical question. Every surface marker registered as confrontation. But when I looked at the substance, the substance was exactly my point, extended further. She wasn’t disagreeing. She was saying the principle I’d articulated isn’t autism-specific at all — which is true and reflects the deeper logic of the framework I work in.
So why the misreading?
Affective realism and the construction of tone
Lisa Feldman Barrett’s theory of constructed emotion holds that emotional experience isn’t received; it’s built. The brain is a predictive organ. It uses prior experience, conceptual repertoire, and bodily state to construct meaning out of ambiguous signals. What we experience as the affective content of a message — its tone, its intent, its heat — is a construction, not a perception of an objective property.
Barrett’s concept of affective realism names this directly: we experience our affective interpretations as objective features of the world. When I read the caps as harsh, that wasn’t a perception of something inherent in the text. It was a construction based on my conventions for what caps mean. In my register, emphasis tracks stakes or disagreement. Caps mean a person has raised the temperature. So I read raised temperature, and my body responded as if I’d been challenged.
For my interlocutor — and for many people in registers I haven’t lived inside — caps function as emphatic agreement. Finally, someone said it. The heat is directed at the field, not at the interlocutor. Both readings are built on real conventions. Neither is the objective reading, because there isn’t one.
What Milton actually argued
This is where the double empathy problem comes in. As Damian Milton originally articulated it, the double empathy problem is not an autistic deficit story softened by symmetry. It’s the claim that mutual understanding requires shared norms and predictive priors. When two people don’t share those, the gap is structural — located in the fit between them, not in either one of them. Neither person is failing to read the other. Both are reading accurately within their own predictive frame and arriving at different constructions because their priors differ.
What makes this exchange a clean case is that both of us are autistic.
As DEP has traveled through clinical practice and popular accounts, it has sometimes been received in a softer form: the autistic person and the non-autistic person each have trouble reading the other, but really, we’re emphasizing that the non-autistic person has trouble too. That reception is still organized around a categorical autism/non-autism axis. Milton’s original argument is more general than that, and Bolis’s dialectical misattunement framework makes the generality explicit. Misattunement is what happens when predictive priors don’t align, and predictive priors don’t align along many axes — not just neurotype.
Different academic and activist registers calibrate heat, irony, directness, and emphasis very differently. Some use capitalization as solidarity. Some use it as alarm. Some use it as agreement so strong it spills over the edges of the sentence. Measured-prose communicators — autistic or not — often don’t use capitalization at all and read it, when it appears, as a register shift toward conflict. Her register and mine are both internally coherent. They both belong to autistic people. They don’t share a convention for what caps mean.
The “double” in double empathy is structural, not categorical.
Why this matters for research
When mismatches like this get folded into autism-specific cognitive theories — theory-of-mind deficits, weak central coherence, mentalizing impairments — what is actually a translation problem gets misread as a perceptual or cognitive one. Barrett’s work suggests there is no perception without construction. Milton’s work suggests there is no construction without convention. Put them together and the methodological lesson is sharp: the unit of analysis is the fit between two predictive systems, not the capability of either one in isolation.[^1]
1. Milton’s framework is sociological in origin; Barrett’s is neurobiological. The two operate at different levels of analysis but converge on the same structural claim — that meaning is constructed within shared (or unshared) priors, not received as a property of the message.
Dimitris Bolis’s work on dialectical misattunement points at the same structure. The mismatch is what we should be measuring. Not whose brain works correctly. Not who failed to read whom. The dyad, in context, with its shared and unshared priors.
I almost answered as if I’d been attacked. The substance had told me she was agreeing, agreeing forcefully, and the surface markers had told me she was attacking. I had to slow down and notice that the conflict between substance and surface was not noise. It was the data. It was telling me her conventions and mine weren’t the same, and that her message was reaching me through a translation layer I hadn’t realized was there.
That’s what I want research on autistic communication to take seriously. Not — do autistic people read tone? That’s the wrong question, and it’s been the wrong question for a long time. The right questions are: under what conditions do any two communicative systems share enough predictive priors to converge on meaning? When they don’t, what does the mismatch look like? Whose conventions get treated as default — silently, by whom, with what effect?
The four-subtype paper is progress only if it’s a step toward this kind of architectural thinking. If it becomes four boxes instead of one, we’ve learned nothing.
The exchange described here is real. Identifying details about the other party have been omitted out of respect for her privacy, and the comment itself is paraphrased rather than quoted. From the perspective of the Evolutionary Stress Framework (ESF), which is a conceptual lens rather than a clinical or medical claim.
Leave a Reply