My experience in testing (and sometimes I would actually see this happen if I was looking at my phone at the right time) was that a camera would generate a “regular” notification (“Person Detected” or whatever) and then a few second later the Descriptive Alert would replace it after the AI had time to do the analysis and generate the narrative. I really liked that it was a notification replacement and not just stacking more information onto the initial message by generating an additional notification (which would be spammy).
I’m enjoying reading your impressions and especially appreciate the Keith Morrison reference. 
