You do realize that is exactly what you're doing, too, right?

The point is to use the LLM tools to generate the best arguments on both sides of the opposite assumptions. And then interrogate the assumptions themselves to see which is more plausible (or wait and see which ends up being the way the world works)