I'm curious about the applications of this to machine learning in public policy. In papers like Big Data's Disparate Impact (https://papers.ssrn.com/sol3/papers

I'm curious about the applications of this to machine learning in public policy. In papers like Big Data's Disparate Impact (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2477899 https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2477899) we see misguided and underbudgeted policymakers using very sketchy and very opaque ML algorithms to, for example, decide who stays in prison and who does not. If we can prove that the "decision problem" of who stays in prison and who does not is undecidable, invariant of the specific implementations of the ML algorithm, this could make a case for stopping such overreaches of ML.