I agree, all concerns are justified - well, nearly but first:
After all, we are talking about artificial intelligence.
What does that mean?
Humans have the technological means to manipulate this intelligence.
How the weightings are placed … well, that … is another question that doesn’t fit in this “paragraph” 
Indeed, I would!
I have autistic traits myself and the following research already exists:
AI robots instruct autistic children how to learn to understand the emotions of “neurotypical” people.
AI does this without introducing “typically human” emotions itself.
However with a great deal of AI-specific empathy 
This is also done using pattern recognition.
Indeed, autistic people and AI learn how to be “empathic” by evaluating patterns and empirical values.
I would trust AI to be more objective, rational teachers!