When I attend tech conferences, people usually know me — if they do — for my work on Cilium and my labs.
That started changing recently. At the last two tech conferences I attended, visitors came up to me and thanked me for my series on autism. That has been extremely encouraging, especially hearing them say they relate to most of what I shared.
But their next question is often: "so how would you define autism, actually?"
It's a fair question. And the honest answer is that the official definition — the one in the DSM-5, the handbook clinicians use worldwide — doesn't really answer it.
The DSM defines autism by its observable symptoms: persistent deficits in social communication, restricted and repetitive behaviors. It's a checklist of what a clinician can see in a room, or what a patient can report about themselves. It describes the shadow, not the object.
This made sense historically. Autism was identified and named from behavioral observation before its neurological mechanisms were understood. The definition was built on what was available — the same way plague was once defined by buboes and fever, before Yersin identified Yersinia pestis in 1894 and medicine stopped defining the disease by its symptoms and started defining it by its cause.
We are still, for autism, in the pre-Yersin era. Not because the underlying neurological differences are unknown — neuroimaging studies have consistently identified atypical subcortical connectivity, different sensory processing thresholds, distinct patterns in amygdala and thalamic function. The neurobiological reality is known well enough to begin building definitions grounded in cause rather than presentation. Clinical practice just hasn't caught up to the science yet.
The goal here is not to settle which neurological mechanism is primary — researchers are still actively debating that. The point is simpler: the definition belongs in that category at all, rather than in the category of observable social behavior.
So the DSM keeps the symptom-based definition. Not out of dishonesty — out of a category error that was reasonable when it was made and hasn't been corrected since. Mistaking the presentation for the condition. The shadow for the object.
We may not yet have a complete mechanistic definition. But an incomplete map of the cause is still closer to the object than a detailed map of the shadow.
The distinction matters because a definition and a diagnostic test are not the same thing. A definition requires knowing what something is. A test requires detecting it consistently. We already know autism is neurological. We know it has a substantial genetic basis. We know many of the brain structures and processes involved. What we do not yet have is enough precision to build a repeatable diagnostic test from that knowledge.
The first bar has largely been cleared. The second has not. Defining autism by its behavioral presentation conflates the two — building the definition from the available test rather than from the available knowledge. That is the category error.
The behavioral criteria remain useful as diagnostic proxies — when used as such. The problem is when they become the definition. One of my children was seen by clinicians for years who dismissed the possibility of autism because he presented well in their office. He was a good boy. Not rolling on the floor. What they didn't see was what happened at home. The proxy was working as designed. The clinicians mistook it for reality.
There's a compounding error here: early support and intervention can improve behavioral presentation. Which means the better the support, the less visible the condition — and the more likely a clinician using a behavioral checklist will miss it. Adjusted behavior gets mistaken for low need. A social outcome gets mistaken for a sufficient one.
Some would argue that if a child presents well, diagnosis is unnecessary. But diagnosis changed everything — not for the clinicians, but for the people around him. Siblings understood. Teachers adjusted. The bullying stopped, not because he changed, but because the people around him finally had a frame for why he was the way he was. That is what diagnosis is actually for. Not a label. A lens.
The consequences extend further. A definition built on observable behavior misses anyone whose behavior has been masked, compensated, or trained away. Which is precisely the late-diagnosed adult population — the people most likely to have spent decades performing neurotypicality convincingly enough to fool everyone, including themselves. My own GP discouraged me from seeking a diagnosis twice. I was diagnosed at 43.
As a suggested definition, for what it's worth: autism is a neurological condition — not a psychological disorder — in which the preconscious processing of sensory input and social information operates with a different filtering baseline, through different mechanisms and different thresholds than the neurotypical baseline. The behavioral symptoms follow from that. They are consequences, not causes.
That definition is stable. It doesn't shift with age, context, or the quality of someone's masking on a given Tuesday.
The checklist shifts. The underlying neurology doesn't.
I'd be curious to hear from researchers, clinicians, or fellow autistics: is there a better definition? One that names the cause rather than the shadow?
SOCIAL SHARE CARD GENERATOR