This is a good example of how ML should be applied in general. I am sick of always seeing this all or nothing approach where people want an AI magic black box that does everything, including wiping their backside. If it can’t do one small thing perfectly it is a failure because there is no way to modify its behaviour as a user and that one thing might be important enough to invalidate the entire approach. What we need is for ML to be applied in very limited scopes to create tools that make humans work more efficiently. Over time these will grow more powerful and general anyway, so we might as well make them useful in the interim.