Robustness score for an opaque model

This invention describes a way to understand why an artificial intelligence (AI) model makes a particular decision. It works by taking a data point, having the AI model classify it, and then finding a slightly altered version of that data point (a 'counterfactual') that would cause the AI to make a different classification. This helps reveal what aspects of the original data point were critical to the AI's decision. The claims specifically focus on finding the closest possible counterfactual, potentially using a genetic algorithm.