The issue isn't the use of DL per se, it's that the resulting NN has independently discovered it can achieve its goal of increasing "engagement" by fuelling negative feedback loops in its test subjects (us). There's blind faith in its effectiveness on the part of its owner, and no oversight.