The idea that you can tell someone’s moral character from microexpressions is hippie woo-woo.
The idea that you can detect the presence of mental disorders from microexpressions is quite valid.
I assume that what you can do is physically possible. Therefore, it must be the latter.
Any time the police train an AI to predict crimes, it becomes racist. It turns out, harassing black people is statistically an effective way to predict crime. Because centuries of harassment, profiling, oppression, slavery, and et cetera have created the socioeconomic conditions to produce crime.
Likewise, if your goal is purely to prevent conflict in an amoral way, you should indeed exclude all neurodivergent people. Autistic people could misunderstand social norms. ADHD people could fail to honour commitments. BPD people could cause drama. NPD people could demean others. Schizophrenic people could imagine threats.
You’re right, it’s mathematically effective to trust the neural network heuristic in your head to tell you which people are trouble. It’s also wrong. It’s discriminatory. You will legitimately solve problems by excluding people who are different, and you will also deny those people an opportunity to show they are more than meets the eye.
I don’t need microexpressions to tell that Trump is a liar. Everything that comes out of his mouth is a lie. Billions of neurotypicals have seen him talk, and hate him. You’re not talking about a skill shared by billions of neurotypicals. You’re talking about a special autistic superpower to read people. I don’t think it’s magic, I think it’s profiling.