Most organizations use a really flawed methodology to hire people in the first place (some combo of phone scree, behavioral, technical screen, team interview). Whether you know the area you're interviewing for or not... you aren't going to figure out if someone knows their shit in 3 hours of talking to them anyway because the only way to know for sure if someone can do something is to have them actually do that thing. Unfortunately most companies don't like this concept (this is not a reference to take home projects...)

Assuming you want to skip actually working with the prospect and you want to try to assess via an interview format, you're looking for one thing: Will this person help your company make money? I.E. will they get results.

While you can never answer that question for sure, looking for human beings that are results oriented is a reasonable first step. Machine Learning as an example is typically a project based type of work. Ask about their projects, how it went, and what was the result. If you talk to enough people about the results of their work, someone will stand out (maybe many people)... then you can move along to what normal hiring managers do anyway: hire who they like the most or pick out of a hat.

The other thing you can do is get expertise in the area.