I recently dug into AI, went quickly into NLU theory, only to perceive it as more or less a better “regular expression”, however I guess there is a need for something better.
I would imagine a user, who can easily make a bot taking a complete medical history for example, where it handles multiple intents, gracefully walks through “topics”, fills various fields of the patient record. Or a bot to test a student understanding of a subject, and provides a score per area.
Making the dialog flow, one by one is ok for simple cases, but making it for the complex ones would make almost unmaintainable flow, taking huge effort to make cases for all scenarios.
Also, I am not sure if the ML algorithms are the best to handle such cases, or if we don’t have sufficient training data.
How about using data from dictionaries or thesauri?
I imagine there are members here who are NLU experts, and wonder if they like to share their experience.
Thanks a lot,