Is There Something Causal Needed to Make Deep Learning Work?

It seems to me there's some kind of tension. Like you said with the semantics. Is there something causal needed to make this actually meaningful? Maybe some notion of temporal consistency and motion is necessary. But I don't know right. It's difficult to say it does not going to work unless you have an existence proof that sorry unless you have a theoretical proof that it cannot work. So far it doesn't exist yet but proving that it cannot exist is much harder. That'll be super cool to have.