Simple theories with good explanatory power are beautiful because, as Juergen Schmidhuber likes to say, beauty is compression. And in this sense, predictive coding theory is a beauty hyperstimuli. Tying together a wide-variety of psychological phenomena, predictive coding captures the essence of the brain’s higher-level functions, providing impressive insight into how our psychology works and how it fails. It’s almost too impressive.
Many researchers have applied predictive coding to better understand autism, and I recommend George Musser’s article for a nice overview. Here I provide the tl;dr version of the article as well as some speculation about why aspies are drawn towards abstract domains like math and programming.
Predictive coding — also known as predictive processing, though there is a distinction — models the brain as a hierarchical Bayesian model that is constantly generating predictions about incoming sensory input. The error in those predictions is fed back up the hierarchy where it is used to update the model’s parameters to make better predictions in the future.
The theory holds that your brain’s predictions are a function of its prior expectations and the incoming sensory data, and it is suggested that autism stems from either a failure to learn good priors or a failure to balance the relative influence of priors and data, putting too much weight on the latter. This means that autistic brains constantly overfit to the data.
Normal brains are able to tune out noise in the environment, but where neuro-typicals see noise, autistic people see signal. They expect precision where there is randomness, causing them to be constantly surprised. In other words, they constantly experience a high degree of prediction error, which can be overwhelming.
This explains why autistic people are drawn to repetitive behavior, why they experience frustration when routines are disrupted, and why they can’t handle stimulating environments. It also might explain why autistic people take things too literally: to not take things literally, you have to rely on priors (what someone probably meant) to sort of trump the evidence (what they actually said).
I’m not sure that severe autism can really have positive benefits, but mild autism such as Asperger’s syndrome might have some upsides that predictive coding can help explain. The high-precision brains of aspies are somewhat unsurprisingly drawn to high-precision domains like science, programming, and math. Hans Asperger himself said, “It seems that for success in science or art, a dash of autism is essential.”
But there seems to be a contradiction lurking here, at least in my mind. Predictive coding views autism as a failure to learn and/or apply priors, which I believe stems from a failure of abstraction. So it should be surprising then that aspies would thrive in abstract domains like math. Being very hand-wavy, I would explain this by saying that the capacity to discover abstract formalisms in a messy world is not the same as the capacity to reason within those formalisms.
The diagram to the left should help clarify what I’m trying to say. I expect aspies to be bad at (1) and (3) but good at (2).
You could also argue that highly abstract domains are vague, not precise, because they are so far removed from concrete observations and they throw away so much detail. It might be hard to understand, for example, how the symbols in an abstract piece of software code might relate to the real world. By this account, you would expect abstract domains to repel aspies.
Yes, abstractions could be considered vague in the sense that they are many-to-one functions that can’t be perfectly reversed (though generative models like our brains seem to do a good job of filling in the details/de-abstracting given the context). But as Dijkstra puts it, “Being abstract is something profoundly different from being vague … The purpose of abstraction is not to be vague, but to create a new semantic level in which one can be absolutely precise.”
Another possible upside to Asperger’s is that aspies may be able to appreciate beauty, in the Schmidhuberian sense, to a degree that neuro-typicals cannot. The more sensitive one is to chaos, the more beautiful the discovery of an underlying pattern or theory that would make the world look more structured, less chaotic.
And then there’s this: