There's two kinds of ignorance which come into contact when people work across disciplines.

In my own work helping ecologists, I see plenty of CS/ML folks who think they'll change the world by throwing a transformer at the problem. (which problem? you think we haven't tried that?) It takes some time and exposure to figure out what kinds of problems you can meaningfully contribute on.

On the other hand, I've met lots (most?) of ecologists who underestimate the impact of looking at their work through CS/ML lenses. Effective automation can greatly improve iteration speed, which ultimately leads to better outcomes than a slow but 'perfect' process. (and, indeed, the 'slow-but-perfect' process may not be sufficiently benchmarked, and not be perfect at all...)

You can do a lot of good by working closely with a practitioner, and identifying the places where they are spending a lot of time doing 'boring' stuff, and finding ways to automate or approximate the outcomes of that boring work. As you work with more people, you'll be able to identify boring stuff that everyone in the field is stuck doing.

So, in short, an excellent goal is to find ways to save people time through bottleneck analysis. Improve iteration speed and you improve the speed at which we can accumulate knowledge / make discoveries / etc. When you're done, it's "just a tool", but beforehand it's a problem holding back discovery.