Any number of ways also includes more tedious. There are already a lot of workers effectively performing QA on ML/AI verdicts (e.g., object recognition, malware analysis).