I started reading "Amazon Unbound" by Brad Stone, and in the part where he discusses the development of Alexa, my laymen self had never considered how difficult it would be to train that thing. I think the term was "far-field" speech recognition, and all the problems that come up when adding in real world variables. People talking from across the room while others are talking, or the TV is going, or as you said, reflections and other materials in the room.

I agree that it is disturbing they tried to use this as evidence.