Automation and social security decision making in the aftermath of Robodebt

Leanne HoSocial security rights review

Daniel Turner, Welfare Rights Centre Sydney   

As artificial intelligence (AI) and Automated Decision Making (ADM) become an ever-increasing feature of our lives – what are the consequences when it is used in social systems to determine social security income support for people relying on it? Or to raise public debts against vulnerable individuals who may have no way to understand why a debt has been raised against them, or meaningfully challenge its legality?

Robodebt, for example, is widely condemned as the worst case of ADM gone wrong in the social security context in Australia. It was described by Federal Court Justice Bernard Murphy as a “massive failure of public administration” as he approved a settlement worth $1.8 billion for hundreds of thousands of people who were wrongly issued with Centrelink debts[1].

Despite this massive failure, Economic Justice Australia member centres see, through their casework, a continuation of ADM gone wrong and perpetuated practices by Centrelink that replicate the conditions for another Robodebt to emerge.

Take Jenny’s* case for example. Jenny is a single parent of three children who had experienced domestic violence before and after separating from her husband. Following her separation, she began receiving the Family Tax Benefit (FTB) and Parenting Payment. Two FTB debts were raised against her amounting to almost $20,000. The debts resulted from an assumed increase in her child support assessment that she had been deemed to receive by an automated decision. However, Jenny had never received this increase as her ex-husband was erratic in his child support payments. His volatility made it difficult for her to pursue him for arrears.

Jenny approached the Welfare Rights Centre in Sydney after she had been unsuccessful in her appeal to the Social Security and Child Support Division of the Administrative Appeals Tribunal (AAT) to have her debts waived, and she had appealed to the General Division of the AAT. We advised Jenny to approach a Centrelink social worker to discuss the possibility of being retrospectively granted a partial exemption from the requirement to collect child support from her ex-husband, which would have the effect of reducing her debt. Unfortunately, the social worker was unaware of the ability of Centrelink to grant such exemptions retrospectively and declined to assist her. We then agreed to represent Jenny in her AAT appeal. Following correspondence and negotiations with the legal representative for Centrelink, Jenny was granted a partial retrospective exemption from the maintenance action test, and her debt was reduced from $20,000 to approximately $2000.

We also continue to see a complete lack of transparency through opaque communications containing little information about how and on what basis ADM decisions are made. This makes it difficult, or even impossible, to appeal a wrongful decision based on an ADM process.

In our session at Economic Justice Australia’s annual conference on 30 August 2022, I will engage Professor Paul Henman, Chief Investigator and UQ Node Leader (ADM+S Centre), in a conversation about AI and ADM – demystifying the language, exploring key concepts, values, and applications in social security and other contexts, as well as looking at the risks and benefits of ADM. Informed from lessons learned from other jurisdictions, the conversation will look ahead to consider what the future holds in this evolving area.


* Name and some identifying details have been changed for privacy purposes.