How to implement the filling of the database of dialogs for further use in the expert system?

Question

Algorithms

It is necessary to make an expert system that will "conduct" a conversation with users according to a certain scenario.
However, the scenario cannot be specified strictly algorithmically.
We have a rough script that we can feed to the experts. Experts will conduct 100, 200, 500, ... conversations with users.
We can feed the resulting chains of dialogs to moderators, who will supplement the database with some machine-understandable tags, flags, and so on.
The resulting database is supposed to be fed to the expert system so that it leads the conversation.
Is this approach correct? In what direction to dig in order to determine the list of machine-understandable tags and the decision-making core?

Answer 1

In the direction of AIML, for example
The best chat bot ALICE www.alicebot.org works on it