The problem is combining multiple people's data for use in ML, and also establishing provenance of different data sets; existing solutions would reveal privacy completely in these use cases.