I share this dilemma too. Just a thought -- I feel less okay with AI processing "data made for humans" (i.e. the images themselves, audio recordings of speech) and more ok with it processing "data made for software" (exif data, shazam logs).