Can AI talk effectively to someone with extremely opposing views?