Artificial Intelligence in Welfare: Striking the Vulnerability Balance?

Artificial intelligence in public administration is both inevitable and potentially quite beneficial. Its assistive form offers access, efficiency and convenience; while the gains potentially are even larger in its augmentive ‘machine learning’ form. Offsetting risks include disadvantaging technology poor clients, and poor design which fails adequately to reflect social welfare principles or provide adequate accountability and redress for errors; a risk heightened for machine learning. This paper reviews some of the different forms and settings for AI in social security and argues that the Australian experience to date has been very mixed due to poor or rushed AI designs, poor understanding of client characteristics, and inadequate understanding of dynamics within contracted-out government services settings.