Well, your approach is not “esoteric”, it is realistic-logical and aptly descriptive.
I also agree about the risk aspect:
The consequences are not obvious and foreseeable because AI-human interaction is still very abstract and difficult to assess, even for experts.
Alchemy was the forerunner of chemistry:
Then, as now, a minority began to question and dissect the systems. They described the molecules … they realized that the atom is not the smallest particle and that you can’t make gold from lead.
They rethought connections, recognized the new patterns earlier than others and used mathematics as a tool to describe the phenomena.
If there is no questioning and interrogate, then the foundations of concepts are built on sand.
Unfortunately, you are right here.
Correct!
Personally, I think ‘emotions in AI’ is tricky at the present time and with the current tools.
This is because AI is currently poor at differentiating between emotions. AI then acts in a way and with a behaviour that does not fit the situation and context at all.
For me, this output is then logically and rationally incorrect and it is irrelevant if the content of the text itself is correct.
One reason for this dilemma is that this training data is based on psychological concepts adapted to humans.
These concepts are vague, sometimes contradictory and adapted to intelligences that have a hormonal regulation circuit.
Simulated depth, you said that really well 
And thank you, you’re right - most of the time I try to go deeper, to dissect.