Using supervised learning to evaluate a result that surpasses human analytic capacity

More often than not, an AI solution exceeds a human's capacity to analyze a situation in detail. It is often too difficult for a human to understand the millions of calculations a machine made to reach a conclusion and explain it. To solve that problem, another AI, ML, or DL algorithm will provide assisted AI capability.

Let's suppose the following:

• The raw data preprocessed by the neural approach of Chapter 2, Building a Reward Matrix – Designing Your Datasets, works fine. The reward matrix looks fine.
• The MDP-driven Bellman equation provides good reinforcement training results.
• The convergence function and values work.
• The results on this dataset look satisfactory but the results are questioned.

A manager or user will always come up with a killer question: how can you prove that this will work with other datasets in the future and...