I appreciate the rigor being asked for in this article, but I wonder sometimes if the typical clinical trial data will go far enough.
"Someone is depressed, is randomly assigned to a therabot, decreases more in a measure of depression than controls, therabot is approved."
But what about longer term effects? What is someone learning or not learning about social relationships? There's an old adage that getting people with agoraphobia to therapy is part of the therapy — so what if someone interacting with a real human would be a lesson in lessening social anxiety or whatever?
I guess sometimes I worry that the protocol - driven nature of modern medicine will spill over into therabots and leave a lot of more complex issues on the table unaddressed.