The key challenges include the lack of sufficient transparency of AI methods and the lack of intelligible ‘reasons for decisions’.
Despite the vast potential of artificial intelligence (AI) to enhance access, efficiency and convenience in public administration, its application in Australia has been hindered by rushed designs and poor understanding of client characteristics, according to Emeritus Professor Terry Carney AO, FAAL.
Professor Carney, a former member of the Administrative Appeals Tribunal, will present a review of Australia’s experience automating social security at “Technology, public law and public administration”, a conference presented by The Australian National University (ANU) Centre for International and Public Law on Friday 1 November.
Professor Carney noted the welfare debt recovery scheme, better known as robo-debt, reflected a “trinity of failings” across basic math, the law and AI design. The system was scrutinised at a recent Senate inquiry following public criticism over automated demands for payment for old debts, despite many people not actually owing money to the government.
“The key challenges include the lack of sufficient transparency of AI methods and the lack of intelligible ‘reasons for decisions’ as well as design barriers preventing speedy access to proper review,” said Professor Carney, who is based at the University of Sydney Law School.
Robo-debt sheds light on other problems for the law around AI and machine learning, he added, including shifting the onus of proof on individuals as well as the power and information imbalance.
“There is also a lack of appreciation of the ‘digital divide’ that results when administration is moved onto apps. This affects the most vulnerable who do not have smartphones or reliable networks,” he said.
“So far, based on the social security experience, it is nearly all ‘blunder’ and hardly any professionally delivered 'agile design' of the kind promised by AI,” Professor Carney said of the situation to date.
His review of various forms and settings for AI in social security is outlined in his new paper titled “Automating Australian Social Security: Boon, bane or just bungled?”.
Other experts speaking at the “Technology, public law and public administration” conference include Federal Court of Australia judge, The Honourable Justice Nye Perram, and technology and law experts Professor Lyria Bennett-Moses (University of New South Wales), Dr Will Bateman and Dr Philippa Ryan (both ANU College of Law), Dr Lachlan McCalman and Dr Tiberio Caetano (both machine learning experts at the Gradient Institute).
Register here for “Technology, public law and public administration”. Registrations close 5pm, Wednesday 30 October.