Two thoughts:

1. data quality is much more important to businesses than fancy neural networks (like those that can manipulate cat images), yet the academic literature has relatively little to say on anomaly detection.

2. While I dislike the term “AI” for the reasons you mentioned, lay-people seem to understand it better than if I say “machine learning” or “data science”. “AI” represents a black box around data science. Instead of killing the term, why not teach what AI can and cannot do?