It's sort of a self reinforcing problem though - a lot of studies trying to understand biochemical mechanisms are taking our existing subpar labels at face value. They generally don't have the sample size to get remotely close to detection of the probably double digit distinct disorders that are making up the "Autism" group. And that is further complicated by natural heterogeneity found even in control groups.

Personally I think coming up with better behavioral characterizations will be an important intermediate step to identifying bio mechanisms, because it leads to much more tractable lines of questioning. For example, we could be quantifying symptoms like rigidity and repetitiveness of movement, sleep quality, sensitivity in different sensory domains, etc. - which should be increasingly feasible to do at scale these days.

How would you go about treatment for illnesses that have no known cause?

There are illnesses where we have a known set of symptoms, a known progression, and known treatment options, but have yet to discover the underlying set of causes. You may be able to reliably diagnose, and treat, without understanding the mechanisms at play.

You're not going to be able to work with the bureaucratic side of the healthcare system without being able to say what they're being treated for. Those treatment options can be approved for one illness and not another, even when they share symptoms, because of known efficacies.

Having a name to call something helps, even when you have no idea about the mechanism at play.

That's a good idea but nobody knows the mechanism for most brain diseases. When they know they say something like "you have a lesion in your right parietal lobe."

Yes but that is why researchers are actively working to change this. We are aware of the problems, but the system we're dealing with is large and complex.

It takes time and effort to learn about the mechanisms, just because we don't know right now doesn't mean we never will.