The genes that build a good engineer can, on a weaker foundation, build a disability.

Picture a family with an engineer father, an aunt who teaches mathematics, and a son with autism who cannot live independently. If autism simply meant less intelligence, that family would be a coincidence. It isn’t one. Many autism-risk variants are positively associated with educational attainment, verbal-numerical reasoning and cognitive ability, and relatives of autistic people often show exactly this technical-intellectual tilt. Yet diagnosed autistic populations show much worse average functioning than that genetic fact alone would predict. How can the same inheritance make the father sharper and the son impaired?

The cleanest explanation is that autism is not a simple intelligence-minus story. It is an imbalance story: the same machinery that raises local cognitive power can make the whole system harder to stabilize.

Tower and foundation

The mistake is treating autism as one causal pathway. At least three streams feed it. Common inherited variants increase autisticness and often increase intelligence or technical aptitude; these are what the father and aunt carry. Rare de novo mutations, new in the child rather than inherited, tend to be bad for general function and more likely to produce severe disability. Developmental insults such as obstetric complications and infection push the same way. Most real cases are mixtures, which is why one diagnosis can cover a nonverbal person with intellectual disability, a brilliant programmer who can’t manage daily life, and an odd but successful academic. They may share a label without sharing the same causal recipe.

A useful way to hold the mixture is tower versus foundation. Intelligence-promoting variants build a taller tower. But a tower needs infrastructure underneath: regulation, integration, social learning and enough stability to keep the whole thing upright. Mutations and developmental insults weaken that foundation. If the tower outruns the foundation, the system does not become superhuman. It becomes imbalanced. The model explains both ends: very intelligent people can be non-autistic because a tall tower is no problem on a strong foundation, and severe autism can appear without giftedness because a badly damaged foundation can’t hold even a short tower. The son may have inherited part of his father’s tower and been given a weaker foundation to hold it.

This fits Autism and Dimensionality, where the problem is not only more capacity but weaker default tuning and fewer robust paths through a larger search space.

Why the numbers mislead in both directions

Two things blur the picture. First, IQ is a rougher tool here than people want it to be. Autistic cognition is often spiky, high in one domain and low in another, while a full-scale score assumes abilities move roughly together. Testing also imports bottlenecks that aren’t intelligence: misunderstood instructions, freezing under interpersonal pressure, refusing the frame. Some of a low score can be interface failure.

Second, diagnosis is a filter. People with autistic traits who can mask or compensate often never reach a clinic, so the diagnosed population overrepresents the cases where adaptation failed visibly, and the father and aunt disappear into the background as nerdy relatives. The clinical category mixes trait biology with institutional visibility.

Neither caveat makes the problem go away, though. Even after accounting for both, as Scott Alexander’s review argues, diagnosed populations still show unusually low measured intelligence while their relatives show unusually high intelligence. That pattern is too structured to wave away. Autism risk contains both intelligence-promoting and intelligence-disrupting components.

Perception without compression

What does the imbalance feel like from inside? One candidate answer, from Bernard Crespi, is that autism reflects an uneven profile in which perceptual intelligence outpaces verbal and mental-rotation support. The exact subtype claim is speculative, but the broader shape is plausible: a person can have exceptional pattern detection, sensory discrimination, memory or systemizing, while the machinery that normally compresses the world into usable defaults stays weak. The result is not merely more perception. It is perception without enough compression. The system sees too much, filters too little, and has to reconstruct consciously what other nervous systems receive as a bundled default. That is why autistic cognition can feel both exact and exhausting.

Intelligence, on this view, is not a fluid poured evenly through the mind. It is a stack of capacities, and the stack can shear. Treating the son as stupid is the cruel mistake, and treating him as hidden genius is the romantic one. Both flatten a high-variance condition with more local peaks, more local cliffs, and a weaker guarantee that the parts add up to a life that works.

So the useful question about any autistic person is rarely “how smart are they?” The question is not only how much horsepower exists, but whether the whole architecture can bear it. That points to the load-bearing structures: sleep, routines, sensory environment, social translation, executive scaffolding, and relationships that don’t require constant masking. The point is not to worship the tower or shame the foundation. It is to make the whole structure habitable.

References: