I’m a little confused by your skepticism. The underlying research is studying what happens given that an interaction reaches a human. Whatever deflection happens before that is irrelevant to the interaction with people afterwards. The paper studies the latter and the results are specific to that set of interactions, which are presumably more complex or ill-suited to deflection.
Hire 100 new people, give them AI -> they do better with AI
Scenario 2:
Hire 10 new people (because AI has reduced demand for humans), give them AI -> they do better with AI
Both of these scenarios are in-line with the research, but are dramatically different when evaluating the proposition that AI helps low skill workers in general.
Why does interaction need to reach a human though? I'm hoping that once the chatbots get good enough, we can give them all the powers of a human representative and get rid of the human.
