While I agree with your sentiment that this is clearly overblown, and perhaps the usage of “AI” these days to describe simple systems in general, your stance is too strong. Markov chains/models have always been considered traditional AI tools. At least it was taught to me when I did my degree in AI twenty years ago, using the first edition of AIMA.

(There’s a whole topic called Hidden Markov Models, and algorithms for learning them using only observed events, like Baum-Welch.)

There are plenty of people doing fun things using texts to build fun predictive models (look at Botnik Studios), but that also doesn’t mean it’s not AI.