I agree, however, causal discovery methods are not totally mumbo jumbo. For example, in a DAG A➡️C⬅️B, A and B would be uncorrelated. But if you condition on C they correlate. This can be tested using algorithms, discovering some (maybe causal) structure.

In this case, how would an algorithm determine whether this is an example of collider bias, or a missed causal relationship between A & B? I could see a utility in using that to flag up potential collider bias, but the methods appear to be used to build DAGs (divorced from theory)