MGMT 8825M Week 5 Case Analysis Example

Debts Raised Against Debts Never Owed: A Case Analysis of Australia's Robodebt Scheme Setting the Column Its Sponsors Counted Beside the Column the Public Record Later Filled, and Arguing That the Uncounted Harm Should Have Been Estimated and Weighed Before Launch

Student Name

Walden University

MGMT 8825M: Ethics of Artificial Intelligence and Autonomous Systems

Instructor Name

Month Day, Year

Debts Raised Against Debts Never Owed: A Case Analysis of Australia's Robodebt Scheme Setting the Column Its Sponsors Counted Beside the Column the Public Record Later Filled, and Arguing That the Uncounted Harm Should Have Been Estimated and Weighed Before Launch

The Mechanism

Between 2016 and 2019 the Australian government ran an automated debt-raising program officially called the Online Compliance Intervention and widely known as Robodebt. The welfare agency took the annual income that the tax office held for each welfare recipient, divided it evenly across the 26 fortnights of the year and compared the result with the income the recipient had reported, fortnight by fortnight, while receiving benefits. Where the averaged figure exceeded what had been reported, the system treated the difference as possible overpayment and sent a letter asking the person to confirm their income, often for periods several years earlier. If the person did not supply payslips or bank records showing otherwise, the averaged figure was used to calculate a debt. The burden of disproving the debt shifted to the recipient.

The arithmetic contained the flaw. Many welfare recipients work irregularly, earning in some fortnights and receiving benefits in others. Averaging a year's income across every fortnight attributes earnings to weeks in which the person earned nothing and was entitled to support. The method therefore produced debts that did not exist.

The Scale Automation Allowed

Before automation, staff investigated discrepancies between tax and welfare records by hand, at a rate of about 20,000 cases a year. A review of the system in its first year (Commonwealth Ombudsman, 2017) reported that the agency expected around 783,000 interventions in 2016 to 2017. Automation did not create the averaging method; it allowed the method to be applied to hundreds of thousands of people without a person checking any individual case before a letter was sent.

Two Columns

The program's sponsors reported its success in one column: interventions issued, debts raised and money recovered. That column was measured, published in agency reports and used to defend the program in Parliament. A second column, the harms to people pursued for debts they did not owe, was not counted at launch. It was assembled later from the Ombudsman's review, a class action and a royal commission (Royal Commission into the Robodebt Scheme, 2023).

The Royal Commission's final report described the scheme as crude and cruel, and as neither fair nor legal. It found that legal concerns about averaging had been raised inside government well before the program ended.

Table 1

The Counted and the Uncounted Columns

The left column reflects what the program reported while it ran. The right column draws on the sources cited in the text.

The Due Process Frame

Citron (2008) argued that automated administrative systems can strip away the procedural protections that due process requires. When rules are embedded in code, she observed, the notice a person receives may not explain the basis of a decision, the hearing may be replaced by an automated response and officials may defer to the system's output rather than examine it. She called for "technological due process": audit trails, notice that explains how a decision was reached and a meaningful chance to contest it before a human.

Robodebt failed each of those tests. Its letters did not explain that debts rested on averaging. Its process placed the burden of proof on recipients rather than on the agency. And human review, which had once examined each case before a debt was raised, was removed from the start of the process. Carney (2019) described the scheme as failing basic guarantees of the rule of law, including the requirement that the state prove a debt before demanding repayment.

What Should Have Been Estimated

The uncounted column should have been estimated and weighed before the program launched. That estimate was possible. The agency knew from its own manual reviews that many recipients had irregular earnings. A sample of a few thousand past cases, recalculated with actual fortnightly payslips rather than averaged income, would have shown how often averaging produced a debt that did not exist. Even if that rate had been as low as one in five, applied to several hundred thousand interventions, it would have predicted tens of thousands of false debts a year.

Weighing that column would have meant pricing three things: the cost of refunds and litigation if the debts were later found invalid, the cost of staff review needed to confirm debts before notices went out, and the harm to recipients, including time, distress and health. The first two can be stated in dollars. The program's eventual refunds and compensation, more than A$800 million combined, give a sense of the first. Restoring human review of each case before a notice was sent would have cost a substantial share of the program's projected savings, perhaps most of them, which may be why it was not done.

What That Would Have Cost

An honest estimate would have forced a choice the sponsors avoided: keep human review and accept a smaller program with smaller savings, or proceed with averaging and accept a known rate of false debts. The first would have cost the government much of the budget savings it had announced. The second, stated openly, would probably have been unacceptable to Parliament and the public. The program was possible only because the second column went uncounted.

Lessons for Organizations

The case generalizes beyond welfare. Any organization that automates decisions affecting people can fall into the same pattern by measuring the outputs it values and leaving the errors to be counted by someone else. A responsible launch requires an error estimate from a sample, a stated tolerance for false positives and a route for affected people to reach a human before harm occurs.

References

Carney, T. (2019). Robo-debt illegality: The seven veils of failed guarantees of the rule of law? Alternative Law Journal, 44(1), 4-10. https://doi.org/10.1177/1037969X18815913

Citron, D. K. (2008). Technological due process. Washington University Law Review, 85(6), 1249-1313.

Commonwealth Ombudsman. (2017). Centrelink's automated debt raising and recovery system (Report No. 02/2017). https://www.ombudsman.gov.au/publications/reports/investigation

Royal Commission into the Robodebt Scheme. (2023). Report. Commonwealth of Australia. https://robodebt.royalcommission.gov.au/publications/report