As an engineer working in the field who has designed both cloud algorithms and on-firmware algorithms, when marketing uses AI, it tends to just be training/data fitting. At best, the most complicated ones tend to be random forests and if any use neural networks, it’s usually just overkill.
The answer to your last question is yes, especially when it’s from raw signals.
Tbf, there are applications from devices that do use deep learning methods but from experience they are not practical except on very edge cases.
In modern common usage both the terms "AI" and "algorithm" are just newspeak for "a computer does something" so combining the two into a single phrase just superlatively multiplies the value, like how using a double negative emphasizes how very much more negative something is. In the middle ages the term might have been "miraculous" and it could also be well served just by sampling Magnus Pike exclaiming "SCIENCE!".
If you hard-code effective learned distributions from a trained model, I suppose that could be described as an 'AI algorithm', even though the final output is a flat algorithm.
Does it mean AI models are used for data fitting? Or clustering?
For data generated in such low scales, wouldnt statistical methods or procedural methods be sufficient or efficient or both?