All the lie detection methods I've seen are based on questionable science with high false positive rates, there's no known method to know for sure if someone is lying and I don't see how adding "AI" on top of it will change that.