> can we actually interpret these models, or are human brains too feeble to understand complex systems without sophisticated models?

I think we will have to develop a methodology and supporting toolset to be able to derive the underlying patterns driving such ML models. It's just too much for a human to comb through by themselves and make sense of.