How the Robo Debt System went so wrong
In traditional welfare reviews, an officer would look at records to identify suspected underpayments, request pay slips from Centrelink recipients, then request them from employers, and eventually use annual earnings data when that didn’t work.
The new online system did the opposite.
Annual income data was automatically used and averaged over 26 fortnight, assuming income was the same for everyone.
If people earn more than fortnightly, depending on the welfare they are receiving, the amount to which they are entitled usually drops, sometimes to zero.
Nonetheless, the robo-debt system assumed that people on fluctuating incomes, perhaps because they had odd jobs to adjust to college or “care” obligations, actually had a steady income.
As a result, this was more likely to indicate that they were entitled to less money than they actually were. Welfare beneficiaries then had to find evidence to dispute a debt that existed solely on account of a miscalculation.
There were other problems too. Versions of robo-debt asked people to simply “confirm” their income, which people did because it was their actual average income, not knowing that it would be used to determine bi-weekly income and take on debt.
The system also combed years of payments, much longer than people were told to keep records, to dispute the system’s false claims.
If their contact information was out of date, some people first heard of an alleged social problem when debt collectors knocked on their door.
The government adjusted robo-debt in response to feedback, particularly a report from the Commonwealth Ombudsman, and Robert defended it on the basis that the core technology of income generation was not new.
“Using ATO average earnings data to determine benefit entitlement goes back decades and decades, nearly 30 years to the time of Hawke and Keating,” said Robert.
He said the Morrison administration is the one cleaning up the system. “Unfortunately, it’s been a long-standing practice,” Robert said on Sky.
State messing around, especially with IT systems, is nothing new. The COVIDSafe app did not live up to expectations and the 2016 online census crashed.
What makes robo-debt stand out is the number of government warnings. There has been extensive media coverage, almost since the plan’s inception, including a prescient article from 2017 that memorably portrayed the plan as a giant robot Santa Claus with a Centrelink logo for its throat.
That year, the Commonwealth Ombudsman published a report that did not oppose averaging but had serious reservations about its use.
At the Administrative Appeals Tribunal, Gordon Legal said it found 76 decisions invalidating robo-debt.
Professor Terry Carney of the University of Sydney, a former AAT member who made some of these decisions and is now a critic of the program, argues that the government’s decision not to appeal to the AAT suggests they agree to the Was aware of failures of the program.
(In the class action settlement, the government did not admit liability for robo-debt or that its employees were aware of the shortcomings in the system.)
Meanwhile, politicians switched through jobs with responsibility for robo-debt. Prime Minister Scott Morrison was in its infancy Secretary of State and announced his expansion as Treasurer. Christian Porter, now Attorney General, followed him. Alan Tudge was Minister of Human Services for almost two years as of 2016, and Robert, the incumbent Minister of Public Services, also holds the Human Services portfolio.
The turning point came with a federal lawsuit in 2019 in which the government admitted that the method of determining income was unlawful. Labor and the Greens, along with charities, led the fight against robo-debt.
That year, Morrison issued a mea culpa in Parliament.
“I apologize for any harm or harm in the way the government has dealt with this problem and anyone else who has found themselves in these situations,” said Morrison.
Kobi Leins, an expert in digital ethics at the University of Melbourne, says robo-debt does not offer complex lessons on the use of AI in government because the program simply expedited a manual process.
It shows, she says, that technology is no excuse for breaking the law.
“These systems can cause damage so quickly and to such an extent that compliance with the law becomes even more important,” says Dr. Linseed.