OK, because we have patterns of behavior, we do this and that, we do not do this and do not do that when we are in a clear state of mind[1]. Now, to make it usable for AI, we have to define what these behaviors are that you do and do not do that makes you be in a clear state of mind. It is not enough to give an exhaustive list, or use heuristics; you have to consider the context as well.

Sure, humans might do it in a more or less accurate way, but could an AI? If it could, would you please tell me how? I really am curious[2], I work at the hospital.

Oh, about working at the hospital: a patient has been deemed aggressive when in fact it is just her voice and looks and whatnot that makes it seem like she is aggressive. She is not actually aggressive in any way, she just looks and sounds as if she were. The moment you have a discussion with her, you would be able to tell that she is not aggressive at all. How would an AI deal with this, if even humans cannot? You also have to consider that these behaviors are heavily influenced by the environment and changes in environment that one has to consider.

[1] Mind you, there are cultural differences at play here as well. There are some behaviors that would consider you "not in a clear state of mind" in one culture, but not in another.

[2] I am also wondering if an AI would be able to determine if a person has dementia based on behavior alone.