Issue 3 / Justice

December 01, 2017
A photo of scabs during the 1917 Australian general strike.

Schoolboys brought in as scabs clean a locomotive during the 1917 Australian general strike. Photo courtesy of the NSW State Archives.

Austerity is an Algorithm

Gillian Terzis

The Australian government recently tried to replace social services with software. What does fully automated austerity look like?

First there are the text messages. Impersonal, incessant, and devoid of context, they reveal few hints of their purpose. The language is so vague you’d be forgiven for thinking it was spam. “Message from the Probe Group regarding an urgent matter. Please call us.” Then, a deluge of phone calls—up to ten times a day, often after-hours—from an unknown mobile number. Chances are you’re still ignoring them.

Should you choose to return the missed calls, you’ll be greeted by an automated voice and placed in a queue. The identity of your caller remains a mystery. When a human operator at an overseas call center at last assumes the reins—whose identity and place of employment, by the way, remains unknown—you’re asked to disclose your date of birth and home address.

Up until this point it sounds like an elaborate phishing scam, but the stakes seem graver. The voice on the other end grows hostile and threatening. The accusations fly thick and fast: You’ve been lying about your income. You’ve been cheating the system. You owe thousands of dollars to Centrelink, the Australian government’s agency for administering social services benefits.

Welcome To Hell

Countless stories like these have lobbed into my social media feeds over the past year. They belong to what is commonly known as the “robo-debt” scandal, which has stirred outrage in Australia in recent months. Diligent reporting from independent journalists like Asher Wolf, New Matilda’s Ben Eltham, and Paul Farrell from The Guardian have unearthed one horror story after another, driving a rising tide of public anger at the government.

Understanding the scandal first requires understanding how welfare works in Australia. Australia is somewhat unique in that the vast majority of its benefits are means-tested. The consequences of this are complex. On the one hand, Australia’s welfare system is narrowly targeted toward low-income earners, giving the government more bang for its buck—welfare provisions make up a much lower proportion of its GDP than other OECD countries. The rationale behind means testing is to encourage fairness in the welfare system, ensuring that people receive what they are entitled to. Anyone who falls afoul of the system’s stringent requirements must pay back their “debt” to the government.

This “debt”—the difference between what you receive and what you are supposed to receive—is the root of the robo-debt scandal. The scandal’s origins date back to 2011, when the Australian Labor Party introduced a data-matching algorithm officially known as Online Compliance Intervention. Its purpose was to compare the earnings reported by welfare recipients to the social services agency Centrelink with the earnings reported to the Tax Office by their employers. Discrepancies would be investigated by Centrelink staff members, who would then decide whether to follow up with the recipient by letter or phone.

Things changed in December 2016, when the government announced that the system had undergone full automation. Humans would no longer investigate anomalies in earnings. Instead, debt notices would be automatically generated when inconsistencies were detected. The government’s rationale for automating the process was telling. “Our aim is to ensure that people get what they are entitled to—no more and no less,” read the press release. “And to crack down hard when people deliberately defraud the system.”

The result was a disaster. I’ve had friends who’ve received an innocuous email urging them to check their MyGov account—an online portal available to Australian citizens with an internet connection to access a variety of government services—only to log in and find they’re hundreds or thousands of dollars in arrears, supposedly because they didn’t accurately report their income. Some received threats from private debt collectors, who told them their wages would be seized if they didn’t submit to a payment plan.

Those who wanted to contest their debts had to lodge a formal complaint, and were subjected to hours of Mozart’s Divertimento in F Major before they could talk to a case worker. Others tried taking their concerns directly to the Centrelink agency on Twitter, where they were directed to calling Lifeline, a 24-hour hotline for crisis support and suicide prevention.

At the end of 2015, my friend Chloe received a notice claiming she owed $20,000 to the government. She was told that she had reported her income incorrectly while on Youth Allowance, which provides financial assistance to certain categories of young people.

The figure was shocking and, like others in her position, she grew suspicious. She decided to contest the debt: she contacted all of her previous employers so she could gather pay slips, and scanned them into the MyGov app. “I gave them all of my information to prove that there was no way I owed them $20,000,” she says.

The bean counters were unmoved. They maintained that Chloe had reported her after-tax income instead of her before-tax income. As a result, they increased the amount she owed to $30,000. She agreed to a payment plan, which will see her pay off the debt in fortnightly installments of $50 over the course of two decades. “I even looked into bankruptcy because I was so stressed by it,” she says. “All I could think about was the Centrelink debt, and once they upped it to 30k, I was so ashamed and sad and miserable,” she says.

As coverage of the robo-debt scandal spread, calls for the government to suspend the scheme mounted. Yet it refused to halt the program until an inquiry by the Australian Senate finally ordered it to do so in May 2017.

Dehumanized Debt

Automation is dehumanizing in a literal sense: it removes human experience from the equation. In the case of the robo-debt scandal, automation also stripped humans of their narrative power. The algorithm that generated these debt notices presented welfare recipients with contrasting stories: the recipients claimed they’d followed the rules, but the computer said otherwise.

There were few official ways to explain one’s circumstances: twenty-nine million calls to Centrelink went unanswered in 2016, and Centrelink’s Twitter account seems explicitly designed to discourage conversational exchange. One source of narrative resistance is, a website run entirely by volunteers that gathers false debt stories from ordinary Australians so that the “scandal can’t be plausibly minimised or denied.”

Over time it was revealed that many of these debts were miscalculated or, in some cases, non-existent. One man I’d read about was on a government pension and saddled with a $4,500 bill, which was revised down months later to $65. Another recipient, who was on disability as a result of mental illness, had a debt notice of $80,000 that was later recalled. A small proportion of recipients were exclusively in contact with private debt collectors and received no official notice from Centrelink at all.

Soon it emerged that social services were a lucrative avenue for corporate interests: this year’s Senate inquiry revealed that some private agencies tasked with recouping debts were working on a commission basis, pocketing a percentage of the debts they had recovered for the government regardless of their validity. (All debt notices issued by private agencies were eventually rescinded after government review in February 2017.)

The methodology of the algorithm itself was riddled with flaws. It calculates the average of an individual’s annual income reported to the Australian Tax Office by their employer over twenty-six fortnightly periods and compares it with the fortnightly earnings reported to Centrelink by the welfare recipient. All welfare recipients are required to declare their gross earnings (income accrued before tax and other deductions) within this fourteen-day period. Any discrepancy between the two figures is interpreted by the algorithm as proof of undeclared or underreported income, from which a notice of debt is automatically generated.

Previously, these inconsistencies would be handled by Centrelink staff, who would call up your employer, confirm the amount you received in fortnightly payments, and cross-index that figure with the one calculated in the system. But the automation of the debt recovery process has outsourced authority from humans to the algorithm itself.

It’s certainly efficient: it takes the algorithm one week to generate 20,000 debt notices, a process that would take up to a year if done manually. But it’s not a reliable method of fraud detection. It’s blunt, unwieldy, and error-prone. It assumes that variations in the data sets are deliberate, and that recipients have received more than what they are entitled to. What’s more, the onus is on the welfare recipient to prove their income has been reported correctly and that the entitlements they have received are commensurate within twenty-one days.

Yet, as many critics have noted, this income-averaging method is porous. It fails to accurately account for the fluctuating fortunes of casual or contract workers, which often results in variations between the two figures. There’s also no way for the algorithm to correct for basic errors in the system’s database. It cannot yet discern whether an employer’s legal name has been used instead of its various business names—it treats them as separate entities, and therefore separate sources of income—or whether conflicting reports are caused by basic mistakes, such as spelling errors or typos. These seemingly small distinctions are ones that only a human could make. It’s no wonder, then, that conservative estimates of its error rate hover at 20 percent.

Fully Automated Debt Slavery

Centrelink’s automated debt recovery program is part of an ongoing initiative to expand the range of essential government services provided to citizens online. In an interview last year with ABC Radio National, the Australian national public broadcaster, Prime Minister Malcolm Turnbull referred to this digital shift as an “important part of the government’s productivity agenda,” and promised it would ensure that “citizens can engage with government on digital platforms as easily and conveniently as they do with their banks or e-commerce vendors.” It’s the kind of market logic that treats citizens as little more than consumers.

Automated debt recovery isn’t confined to Australia. In December 2016, The Guardian reported that the Michigan Unemployment Insurance Agency’s automated compliance system wrongly issued debt notices to over 20,000 welfare recipients—93 percent of recipients in total—from October 2013 to August 2015. Those who were accused of insurance fraud lost access to their unemployment benefits, and even had their taxes garnished to pay substantial fines. At the same time, the state’s governor had earmarked the funds raised by the compliance system to balance the state’s budget. It’s fitting, if a little uncanny, that the Michigan Integrated Data Automated System is also known as MiDAS.

These kinds of algorithms are badly designed, but their underlying ideology is more sinister still. Leaks from Centrelink staff members to the press revealed an underfunded government department struggling with depleted morale. Reports revealed that employees were sent memos outlining performance targets within the organization, which pitted employees against each other in a battle to process the most debt notices.

In retrospect, the previous year’s Australian federal budget contained a telling omen: it forecast that $4.5 billion in overpaid entitlements could be reclaimed. It makes sense, then, that the dragnet was deliberately sweeping—its purpose was to balance the books. By the same token, the glitches in the system—from the difficulty of contacting Centrelink, the outsourcing of 43 percent of its debt recovery to corporate interests, and the kinks in the algorithm itself—are by design. They’re features, not bugs, aimed at discouraging formal complaints, raising money, and making the system as punitive as possible.

We shouldn’t be surprised. There are plenty of conservatives for whom welfare is a societal wound that demands excision. No wonder their prescriptions seem so cruel. How else to describe actions that coerce the poor into the labor market at any cost? The infliction of force is inscribed in policy.

Alan Tudge, the embattled Human Services Minister who implemented the aggressive recovery strategy, defended the program by suggesting that Australia’s social spending had been too lavish: “We must face the reality that in our desire to be a generous and caring society we may have reached a point where we have taken our good intentions too far.”

The Patron Saints of Dole Bludgers

It’s 1996. A Current Affair, a tabloid news show, is pondering the plight of Australia’s unemployed. Cameras skim the snaking queues and drab surrounds of a suburban Centrelink office. The narration is at once incredulous and contemptuous. Why don’t these people have jobs? Aren’t they trying hard enough?

The specimens in the show’s lurid anthropological study are the Paxton family, who turn out to be luckless in more ways than one. They live in St Albans, a working-class neighborhood on the outer edges of Melbourne’s sprawling suburbs. It’s afternoon and sixteen-year-old Mark is in bed. The cameras linger on Mark, slumped in repose, while his older brother Shane tries in vain to coax him out of bed. Mark’s nonplussed, and not having a bar of it. “I’ll look at the clock at 9, 10 o’clock, and think, why do I have to get out of bed?”

For another journalist, Mark’s helplessness might have been a hook for an exploration of the psychological and social ramifications of long-term unemployment. But this wasn’t going to be that story. Instead, the program set about cultivating the myth of the indolent poor, a mythical underclass that lounges around, mooches off the system, and has the temerity to spend society’s hard-earned money on booze.

From the get-go, the dog-whistling is slimy and conspicuous, demanding resentment and scapegoating at every turn. There’s a barely contained glee in its whipping up of thinly veiled animosity and malice against the unemployed. From a ratings standpoint, it was a humiliating but high-rating spectacle: millions tuned in to watch the demonization of the Paxton family—specifically its teenage children—as they refused to succumb to the platitudes touting the dignity and virtues of low-paid labor.

Over a period of two weeks, A Current Affair’s camera crew harassed the Paxtons at their home, airing their misfortunes on national television for six consecutive nights, stoking the embers of middle-class rage. Before long, the Paxton teenagers were anointed the “Patron Saints of Dole Bludgers”—people who are on benefits but are believed to be uninterested in seeking work—by a tabloid newspaper columnist, referred to as “putrid” by another, and ordered in a hectoring editorial to take jobs “for the sake of all our children.” Some politicians even called for an inquiry into the legitimacy of the family’s welfare benefits.

The Deserving Poor

For all of Australia’s egalitarian chatter, the issue of welfare and who deserves it has long stuck in the nation’s craw. It prompts moral righteousness among the Right, which has couched cuts in social services as fiscal reform, while progressive technocrats tittered about the best methods to refine means testing. There’s even a tendency by some commentators, perhaps intoxicated by the inoffensiveness of liberalism, to critique these cuts without any reference to ideology at all. They’re presented as a symptom of poll-driven politics, rather than something more pernicious.

Much discussion of social inequality and class divisions in Australia treats these issues as phenomena that occur elsewhere. Home ownership is seen, somewhat uncritically, as an inviolable fixture of the Australian dream rather than a tax racket for the rentier class. Pouring scorn on “dole bludgers” and shaming welfare cheats has long endured as a national sport. Anxiety about welfare dependency persists, even though it is at historically low levels.

In June 2017, the government released a report of the nation’s top ten suburbs for welfare non-compliance—or “bludger hotspots,” to use the parlance of a News Corp tabloid. These are places where recipients failed to meet the conditions for receiving welfare, which include attending interviews and appointments. It’s difficult to see such a move as anything other than a way to shame and stigmatize these communities.

The political language used to talk about welfare reveals the nature of a shifting relationship between citizens and the state. Before World War II, unemployment was seen as an affliction of the “undeserving poor.” It was a social and collective responsibility, but it didn’t require government intervention. In postwar welfare societies, unemployment assumed an explicitly political dimension through the creation of social safety nets for the disadvantaged and economically vulnerable.

As Australian media theorist and academic Alex Griffin writes, this “conception of citizenship entailed the state protecting citizens from unemployment, with social citizenship defined as the ‘right’ to assistance.” But the ascendance of supply-side economics in the 1970s recalibrated the discourse. By framing unemployment as a problem of supply rather than demand, joblessness was presented by conservatives as a moral choice rather than a symptom of market failure. Instead of receiving welfare assistance, the argument went, the poor should bootstrap themselves out of their poverty.

This rhetoric of self-reliance prevailed through the mid-1990s and beyond, in which policies of “mutual obligation” were introduced by the conservative government led by John Howard. As Griffin observes, these policies ensured “eligibility for income support was reconfigured, newly coupled with requirements.” These included “obligatory participation in prescribed employment assistance activities, lower benefit payments and shorter benefit periods, means testing, and punishments for failures to attend job interviews.” And the slightest failure to follow these mandates led to a loss of income.

The bureaucratic language surrounding these benefits—“Work for the Dole”—marked welfare for the economically marginalized as a provisional rather than enshrined right for citizens. It’s support with strings attached.

Yet the irony of stigmatizing welfare recipients is that better-off Australians are major beneficiaries of social spending. The Australian writer Tim Winton notes that the country’s middle class has “an increasing sense of entitlement to welfare,” which is “duly disbursed largely at the expense of the poor, the sick, and the unemployed.” These include tax concessions on contributions to “superannuation,” which are funds designed to help Australians save for their retirement. Such concessions are distortionary: they’re levied at a flat rate of 15 percent, rather than at a progressive rate according to one’s income, which means their benefits are reaped overwhelmingly by the rich.

The Australian Bureau of Statistics calculates that nearly one third of these concessions are claimed by the top 10 percent of income earners in Australia. Then there are policies like negative gearing, a tax concession that allows you to claim a deduction against your wage income for losses generated by any rental properties you own. (Australia and New Zealand are the only countries in the world to hold such a policy.) In addition, Australian homeowners are entitled to a capital gains tax discount of 50 percent once the property is sold.

Critics have argued that the combination of these two policies only serves to fuel investor speculation, entrench housing unaffordability, and lock first-time home buyers out of the market. But it’s easier to attack the poor than to tax the rich.

Shredding the Social Contract

Unfortunately, the fallout from the robo-debt scandal has not deterred the Australian federal government from continuing to use technology to impose “personal responsibility” upon its citizens.

In May 2017, Centrelink announced that it would expand its cashless debit card scheme for welfare recipients living in areas with high levels of drug and alcohol abuse. The scheme quarantines 80 percent of welfare payments to the debit card, which contains software that blocks purchases of restricted products such as alcohol, drugs, or gambling-related goods at the point of sale. It’s already been deployed in two towns (Ceduna in South Australia and East Kimberley in Western Australia), both of which have a sizable indigenous population. It’s a dystopian vision of welfare—one that is discriminating, inhumane, and corrosive.

The creation of robo-debt is a logical, if painful evolution of the neoliberal unraveling of the Australian welfare state, in which social services were subject to savage cuts over the past few decades. Means testing is part of the problem, as was the decision made in the mid-1990s by the Australian Labor Party government—and continued by successive conservative and progressive governments—to bring in private employment agencies to help job seekers gain employment. What was once the responsibility of the state gradually became a lucrative opportunity for private firms to capitalize on a market that previously didn’t exist.

These outcomes weren’t inevitable. The bipartisan embrace of means testing has meant that a truly progressive policy for welfare—that is to say, universal, unconditional, and rooted in a model of social citizenship safeguarded by the state—has not been articulated by policymakers on the left. What exists is a perversion of the social contract in which government shirks its responsibilities for the social well-being of its citizens in favor of serving the interests of capital.

Means testing creates a political climate in which it is increasingly palatable for cuts to be made to the most vulnerable. By contrast, universal benefits are much harder to cut. The few benefits that are universal in Australia—such as Medicare, the government-funded health insurance scheme—are vigorously supported by Australians across the political spectrum. (An attempt by the federal government to introduce a copayment for doctor visits was rescinded after public outcry.)

There’s an opportunity, surely, to make the welfare state more humane, more dignified, and more equipped to address the needs of all Australians. It’s tempting to think of what could have been—how technology could have been used to make social services more transparent and fair, not to mention more accessible to those who depend on them.

What the scandal shows is that the neutrality of technology is a fallacy. A tool is only as good as the politics that underpin it. It’s not an accident that the Australian government’s attempt at algorithmic governance was inhumane. It was a defining feature of its design.

Gillian Terzis is an Australian writer living in San Francisco.

This piece appears in Logic's issue 3, "Justice". To order the issue, head on over to our store. To receive future issues, subscribe.