Legal issues arising from the use of artificial intelligence in government tax administration and decision making

Abstract

This paper examines the issues that arise when government administrative decisions, such as those relating to taxation and social security, are aided wholly or partially by artificial intelligence (AI). As we move further into the digital age, AI is increasingly adopted by the public sector to achieve greater accuracy of analysis and resource efficiency, but it is evident through instances such as the ‘Robodebt’ debacle that there is room for (non-human) error. In administrative law, such decisions can be reviewed judicially or on the merits of the case; however, the difference between error caused by a human delegate and an automated delegate is that we are not yet able to comprehend the reasoning steps undertaken by automated decision makers. This paper highlights the arising need to define what constitutes a decision (and a decision maker) for administrative law purposes, and addresses the implications of when decisions made predominantly by AI are erroneous or cause harm to taxpayers.