The popular press likes to call anything ML related "AI" since presumably that's what they think the public wants to read about.

In any case, the percentage of the population who knows the difference between a transformer, or diffusion model, or a bespoke protein folding model is going to be tiny, so calling it all "AI" does make practical sense despite being a bit misleading (it's not all ChatGPT).

Just to be clear though, the "AI" (AlphaFold) isn't being used to visualize the protein - it's used to guess the 3-D structure of the protein (via folding rules), which can then be visualized.