Hm. I have not founded or sold a machine learning company, or even studied the field. But either of those quantities fully determines the other, so I'd really hope that, if they're part of the model, they'd be exactly equivalent to a learning system.

Do you mean for measuring goodness of fit between a prediction and the known truth in training data?

If you want to assess the presence of knowledge, give a test rather than an interview. The interview is just a way to make it easier for the con artist to snow you.