Much more than that. Insofar as we are talking about public law, I'd argue that ML for granting welfare benefits, calculating taxes, loaning books from a local

Much more than that. Insofar as we are talking about public law, I'd argue that ML for granting welfare benefits, calculating taxes, loaning books from a local library, ticket reserving for public transport, and whatever else goes on in mundane public administration should be open source or at least transparent, including with respect to training data and basic performance metrics like ROCs etc. Note that the argument about trade secrets, by definition, cannot apply in this context, although a practical problem is that public administrations have outsourced their stuff to the private sector (sometimes with catastrophically bad outcomes). It will be interesting to see whether people will increasingly challenge public sector algorithms through FOIs in the future, as I am pretty sure they'll do.