You are talking about AI outright failing to get anything remotely close to the correct result. That's a completely different argument and has nothing to do with non-determinism. Things can be deterministically wrong too.