The question is, do the researchers use the ML/AI system to discover and identify likely text - which could be great - or do they use its output conclusively?

The question is, do the researchers use the ML/AI system to discover and identify likely text - which could be great - or do they use its output conclusively? In fairness, the paper may answer that question but I don't have time now ...