PROJECT OVERVIEW
Digital governance, algorithmic decision making, and administrative burdens: The Australian Robodebt program
Highlights
From 2016 to 2020, the Australian government operated an automated debt assessment and recovery system, which became known as “Robodebt,” to recover fraudulent or overpaid welfare benefits. The stated goal was to deliver $4.77 billion in savings through debt recovery and the reduction of costs associated with public service administration.
However, the algorithm at the heart of Robodebt, and the administrative policies surrounding it, created a system of wildly inaccurate assessments, fraud accusations, and administrative burdens that disproportionately impacted those with the least resources to handle them.
The Robodebt case raises questions about the automation of key public services, the policy goals that automation addresses, and the administrative burdens that can be reduced or multiplied in the process.
Overview
From 2016 to 2020, the Australian government operated an automated debt assessment and recovery system, later dubbed “Robodebt,” to reduce government debt by recovering fraudulent or overpaid welfare benefits. The goal of Robodebt was to use technology and administrative data to detect fraud.
The Robodebt program used an algorithm that averaged annual beneficiary income across biweekly periods and compared them to the actual biweekly income reported by clients to welfare agencies to receive means-tested benefits. If a discrepancy was identified, it was flagged as potential fraud. However, the algorithm was based on a case profile – an individual in stable, consistent employment – that applied to only seven percent of welfare recipients in Australia.
If individuals reported lower income in a biweekly period than the biweekly average of their annual income, this was treated as evidence of overpayment and possible fraud. The assumption of income stability is reasonable for those with salaried jobs, but it was a poor assumption for welfare recipients.
The Robodebt scheme incorporated a second faulty assumption that compounded the first: default judgments made in the absence of a response from the welfare recipient reflected that they, in fact, accepted they had been overpaid.
By the time the program was scrapped, nearly 470,000 Australians had received letters accusing them of welfare fraud. Many of the most vulnerable Australians – those who had received benefits or pensions for years at a time and those who had experienced chronic underemployment – were sent notices that they owed up thousands of dollars. Recipients reported feelings of stress, trauma, and lack of trust in government services.