Ask HN: What would make you trust an AI's assessment of your understanding?

Creating an AI tutor that won't let you progress without having learned anything — not just completing a test. You train, are diagnosed for any problems in understanding, and there is now a live voice challenge where you need to verbally explain your response to an AI, and then it drills down with additional questions.

The voice feature was created by me under the suspicion that it is way more difficult to pretend to have understood something while talking to someone rather than choosing a proper multiple choice answer. But to be honest, I don't know if that is a valuable idea or just another sign of my attachment to the most complicated thing I created.

So, the question I want to ask seriously is: What would help you to trust an AI's diagnosis of your understanding? And is it even worthwhile, or should I stick to the solution of doing spaced repetition and trusting myself?