I think I'm going to do something along those lines. I'm going to make a system to alert a player when a character's statement is suspicious (the AI has a sort of built-in polygraph) and then give the option later on to choose which suspicious statements to question. I haven't decided yet if I'll then require a player to present proof that a statement was a lie, or if just questioning the right statements will be enough. I want to see how complicated it turns out to be to implement.