A new approach toward social licensing of data analytics in the public sector.

Governments using data analytics will be increasingly drawn into creating social licence for these applications. The need will be greatest where two conditions are present. First, where data analytics are used to predict rather than describe or prescribe. Second, where such prediction is used by governments when exercising their coercive powers. Two examples of using predictive risk modelling are identifying children at risk of neglect and abuse; and assessing recidivism risk in the criminal justice system. Each has drawn criticism, precipitating discussion around the requirements for social licence. Much of this discussion focusses on transparency as both an intrinsic virtue and an instrumental virtue for achieving social licence. The paper contends that transparency is unachievable as an intrinsic virtue for such purposes and that its conceptions as an instrumental virtue fall short of that required for users and subjects as well as the public to have "sufficient to approve or disapprove of the algorithm's performance". The conclusion is that unconventional democratic forms, including deliberative and direct democracy, are likely to prove more successful than representative democracy in establishing that licence and thereby realising more fully the potential contribution of data analytics to better government. [ABSTRACT FROM AUTHOR]