Fascinating that they say so bluntly that the end goal isn't a machine that performs well at a task, but warm fuzzies for the people that have to specify the tasks for that machine: "Turning aerial dogfighting over to AI is less about dogfighting, which should be rare in the future, and more about giving pilots the confidence that AI and automation can handle a high-end fight."

Reading further, it's also interesting that they're looking not into how legible the end results are, but how legible the system's behavior during training is.