I think people are hesitant to call ML “statistical modeling” because traditional statistical models approximate the underlying phenomena; e.g., a logarithmic regression would only be used to study logarithmic phenomena. ML models, by contrast, seldom resemble what they’re actually modeling.
We used to distinguish AI as automatically / programmatically making a decision based on an ML model, but I'm guilty of calling it AI for wow factor, lol.
Now I have to be careful because AI = LLMs in common language .