Personally, I'd rather have the people working on AI be the ones who are aware of the potential theoretical pitfalls, rather than someone who is not aware of any pitfalls and rushes full steam ahead.