Public administrators are increasingly confronted with difficult questions about the (0) ALGORITHMIC (ALGORITHM) systems now deployed in welfare, policing and education. The promise of such systems is a kind of bureaucratic efficiency no human process can match; the risk is that they obscure 1ACCOUNT and entrench existing prejudices. Designers insist that their models are 2BIAS, yet the training data on which those models depend is drawn from a world that manifestly is not. Campaigners have therefore made 3TRANSPARENT a central demand, arguing that citizens have a right to understand how consequential decisions about them are reached. The rushed 4DEPLOY of such tools in sensitive domains has prompted regulators to 5SCRUTINY them with unusual vigour. Legal 6PRACTICE now find themselves in uncharted territory, advising clients on questions for which statutes have yet to catch up. Establishing clear 7LIABLE when an automated decision goes wrong has proved particularly knotty, as responsibility is often diffused across developers, vendors and deploying authorities. Whether meaningful independent 8OVERSEE can be institutionalised before these systems become indispensable is perhaps the defining governance question of the coming decade.