This is probably one of the best applications of AI in science in terms of impact so far. I can't think of any other problem with the same potential impact.
AlphaFold is the best counterpoint to tech cynics.
One of the largest public tech companies in the world funded a multi-year scientific project, executed the research flawlessly and moved forward an entire scientific field. They then went on to openly release the code _and_ data, working with a publicly funded organization (EMBL-EBI) to ensure researchers across the globe can easily access the outputs.
I'm not arguing that every tech company is a net positive for humanity. Google itself isn't perfect. Google + DeepMind is setting a bloody high bar though.
This is definitely one of the most exciting spaces in AI right now. Another somewhat-related startup is PostEra (medicinal chemistry for drug discovery via AI) https://postera.ai/about/
You are right and when thinking about it I can see 2 problems which I hope in the future can have even more impact:
1. Using AI to determine the most efficient methods of doing mathematical expressions, transformations and computation algorithms - division, square root, maybe traveling salesman - these which take relatively high amount of CPU cycles to compute and are used everywhere. If inputs and outputs can be assigned to it, AI can eventually build a transformation which can be reproduced using a silicon.
2. Physics phenomena in general, not only organic protein, can be measured and with sufficient ability to quantize them to inputs and experimentally obtained outputs to train the network, we could in theory establish new formulas or constants and progress the understanding of the Universe.
jarenmf said "in science" - but it is an interesting question how much automated translation has helped scientists translate papers from other languages.
EDIT: grammar