What AI can learn from AI in video games?

I was asked this question when watching a youtube video of the same title by the youtube creator Mark Brown. I urge you to watch the video as this article will be mostly based on that video. The first point he makes is that good ai allow the player to cheat. He gives examples like in Skyrim the player can put a bucket on top of a shopkeepers head and rob them. or in farcry that only a few enemies are allowed to shoot at once.The second says good ai tell you what it’s thinking, I believe this is an important statement for machine learning as right now we lack the AI(ai in the normal sense, not the video game) can’t tell you what’s its thinking but solves the problem at hand perfectly. This is what many experts call the black box problem. In the video, mark gave the example of barks where NPCs give short vocal clips to give the idea to the player that the user what the ai is about to do. When doing machine learning obviously you won’t have the model gove you vocal clips, but it will be nice to find a way for the model to give a prompt where it’s heading to solve a problem, I know for evolution models the most fitness model will be more likely to spread its features so the user can have some idea where the model is going to solve the problem. Next in the video mark said what makes a good ai in a video game is predictableness, the reason for this is because when playing the ai should be consistent so the player can develop plans against the ai which allow the player to play the video game with intentionality. In machine learning, you want the same for the model you are training it’s no good to have an unstable loss chart as that means the model is still mostly winging it. Or having a high prediction rate as the results given will be consistently right with some errors. The next point in the video he says good ai can interact with the game system. In the video they gave an example of in Zelda breath of the wild, enemies set their clubs on fire, kick bombs away and even throw other enemies at the link. In machine learning, they are many examples of this. In another video from two-minute papers which gave 4 examples of the ai worked unexpected ways in the environment. In one example it showed a spider-like robot to walk using none of its legs, the robot prompted to flip upside down and use its elbows to move. The difference between the machine learning example and the video game example is that the video game example the ai is purposely scripted to complete those actions but giving the illusion to the player that the video game ai is smart. The machine learning example the authors were generally surprised as those events are not scripted in.