We may simply lack the apparatus to fully comprehend social science. Neural networks as they stand might be the key, but they could also be a mere stepping stone, such as how Aristotle's belief of the world being formed by a handful of basic "elements" was a sufficient and necessary bridge to our contemporary understanding of the atomic and subatomic universe.
It's probably not a good idea to take social science findings too seriously, but that doesn't make it bunk science. Just...imprecise.
The bunk level of a field that claims to be a science can be determined from its predictive power. The predictive power of almost any social science is very, very low.
Even if one's findings turn out to be wrong, that doesn't dismiss the necessity to study the unknown. Would you have given up on space travel after the first few rockets exploded on the launchpad?
Ironically, my parent post is being heavily downvoted for stating this opinion.
What other method is there? Studying a rabbit's entrails?
Others are less so, which doesn't make them any less useful, like frontier thesis: https://en.wikipedia.org/wiki/Frontier_Thesis
And on top of all that, empiricism isn't the only method for discovering truth!