AustLII Home | Databases | WorldLII | Search | Feedback

University of New South Wales Law Journal Student Series

You are here:  AustLII >> Databases >> University of New South Wales Law Journal Student Series >> 2023 >> [2023] UNSWLawJlStuS 6

Database Search | Name Search | Recent Articles | Noteup | LawCite | Author Info | Download | Help

Nemec, Michelle --- "Robodebt Illegality And How Expediting Automated Decision-Making Failed To Take The Bull By The Horns " [2023] UNSWLawJlStuS 6; (2023) UNSWLJ Student Series No 23-6


ROBODEBT ILLEGALITY AND HOW EXPEDITING AUTOMATED DECISION-MAKING FAILED TO ‘TAKE THE BULL BY THE HORNS’

MICHELLE NEMEC

I INTRODUCTION

This article examines the discredited and defunct Robodebt Scheme (‘Robodebt’) as a government automated decision-making (‘ADM’) system, and what can be learnt about the process of using ADM in the public arena, particularly the social welfare system from technological, regulatory, and legal perspectives. The article scope is confined to Robodebt domestically, to exemplify the potential benefits or detrimental outcomes and risks that can eventuate, involving the Australian Department of Human Services (‘DHS’) (now Services Australia) the operators of the ADM system and the welfare recipients the subjects. The discussion about Robodebt focuses on ensuring a system remains accountable and responsible, as Robodebt provides a prime example of what happens when accountability is lacking, and the rule of law (‘ROL’) is threatened.[1]

A literature review demonstrates Robodebt provides the impetus for a wide variety of articles across disciplines including accountancy, technology, law, and politics. This article is of value given the interest ADM and artificial intelligence (‘AI’) presents as, “what was once inconceivable, that a complex decision might be made [about] without any requirement of human mental processes is, for better or worse, rapidly becoming unexceptional”.[2] The controversy Robodebt and ADM has caused, prompts questions about how to do things better in the future. This is complex, involving multidimensional factors, and in fairness, only a few issues can be raised here due to the relative brevity of this article. Furthermore, ADM and AI are emerging areas of technology, meaning the impact and potential of their application in the public sphere should not be underestimated. Moreover, this article provides an entry point for evaluation about ways to avoid a future calamity involving ADM use by the Government as the technology continues to evolve.

The article structure is first to discuss the background to Robodebt, then to explain ADM, scoping out a definition and the boundary specifications used within this article, prior to using different examples to clarify how ADM is applied here. Second, a question is posed about why and how Robodebt fell short on accountability, thereby failing to provide responsible government and transparency exacerbated by the failure of available safeguards “to take the bull by the horns” (activating all measures of safeguards to maintain government accountability while addressing the financial imperative). Third, lessons are examined about what can be learned through the Royal Commission (‘Commission’) and using ADM. Finally, a conclusion with brief recommendations for further research are asserted.

A Background to Robodebt

The chronology of events from the July 2016 launch by the DHS - Centrelink of the online compliance intervention (‘OCI’) system for raising and recovering debts, involves much about failure and neglect, rather than “taking the bull by the horns” to deal with the complexities prudently and judiciously. This “layered failure” includes why the OCI system was doomed to fail from its inception on legal grounds, moving onto the failure of the Government to act on legal advice reported prior to the Commission,[3] and the failure of the economic imperative involving thousands of complaints and the harm[4] caused to people, leading to the Senate Inquiries and the Commission itself (hearings concluded on March 10 2023).[5] As a flawed system, the government’s insistence on pushing forward with it can be likened to “launching a regulatory assault on its own integrity”.[6]

Robodebt was introduced with Government zeal, demonstrated by Alan Tudge’s (the then Minister Human Services) startling message, “We’ll find you, we’ll track you down and you will have to repay those debts and you may end up in prison”. [7] This was stated in response to claims many welfare recipients received fraudulent claims, and the program administered by Centrelink within the DHS was costing the Government vast sums of money unnecessarily. Reducing the national debt has long been an objective of Australian political parties and is regarded as essential to political points scoring. One debt reduction strategy is to reduce and recoup welfare support overpayments. Australia has a controversial history of “demonising” the long term unemployed.[8] In the Robodebt saga, the tone and character of the Government was questionable as Prime Minister Scott Morrison met with his top officials, expressing a desire to be the “welfare cop”,[9] demonstrating arrogance and near “cockiness”. These factors combine to create a backdrop for the “perfect storm”.

These sentiments and approaches are more about acting hastily and without adequate evaluation than taking a responsible approach which could include gaining fully informed feedback and then acting on it. Taking quick steps to get the Government “out of the red” and asserting power and control, rather than being definitive about Robodebt’s legality and acting irresponsibly are completely at odds with valuing fairness and reasonableness, also demonstrating a lack of integrity. Modelling integrity and building a sense of trust are part of the glue for upholding responsible government and democracy. Further, integrity and trust are two foundational relational values which are important for establishing a relationship in any context, or any type. In this scenario the welfare citizens were completely dismissed, and they were compromised.[10] Trust is necessary for cooperation and fulfilling the social contract in democracies.[11]

The OCI system employed automated data-matching algorithms of historic individual’s benefit payment records and income reported to Centrelink with past income tax returns and information at the Australian Taxation Office (‘ATO’), identifying discrepancies between these records. This system reduced human investigation of the discrepancies, with the automatic calculation of overpayments for individuals based on a simple algorithm averaging earnings across the relevant year, followed by raising a targeted individual debt involving some 700, 000 people.[12] The system disregarded the timing of paid earnings and their amount, despite social security payments being calculated fortnightly based on the earnings for each fortnight.[13] Significantly, Robodebt breached the principles of ethical administration and legality by shifting the burden of proof, meaning that the individual had to provide evidence that they had ‘not’ been overpaid, whereas previously the government had collected detailed records from employers, thereby diminishing the work needed to examine individual’s records thoroughly.[14]

Such was the lure of economies of scale from a political fiscal perspective that ADM offers, it was prioritised over the human factor, and the need to take responsibility for ensuring debts were owed by the subjects.[15] The OCI’s expeditious and arbitrary implementation failed to factor in unique individual circumstances. The Government’s aim was to recover 1.7 billion over 5 years in overpayments[16] from welfare recipients and to create a system to curtail extraneous labour and reduce costs. However, schemes such as Robodebt need to stand up to public scrutiny and provide suitable remedial avenues backed by structural procedure and access to do the same. Robodebt did not, instead, failing to deliver the cash flowback in a dramatic way with a class action launched in September 2019.

The Government conceded in May 2020 it would refund $AUD720 million to more than 400,000 individuals for unlawful debts. In November 2020, a Federal Court settlement was reached whereby the Commonwealth agreed to repay the $720 million in debts already collected, to drop further claims of $AUD398 million and to pay $112 million in compensation to approximately 400,000 eligible individual Group Members, including legal costs.[17] This predicament saw the Government not only undoing the promise of increasing reducing costs but “back pedalling” in a not so dignified manner with an accompanying disintegration of public trust, as well as a loss of productivity and internal efficiency.

An extension for the delivery of the Commission’s report was announced on 16 February 2023 with the report now to be delivered on 30 June 2023. It will be a comprehensive report dealing with several key issues. These include who was responsible for Robodebt (establishment, design, and implementation) and why it was deemed suitable and needed to begin with, considering issues and concerns such as fairness, transparency, and legality. Other matters under scrutiny include how concerns raised about Robodebt were managed by the Administrative Appeals Tribunal (‘AAT’), harms, and costs and what is needed to avoid other similar failings in public administration.[18]

B Definition of Automated Decision-Making

ADM involves using technology to automate a decision-making process, using pre-set logical parameters to perform actions, or make decisions – in whole or part without the direct involvement of a human being at the time of the decision making.[19] It can be inflexible and rigid through to reflexive and adaptive, if for example, it includes an option for human decision-making or the final decision to be made. ADM systems range from traditional rules-based systems, such as a system calculating a rate of payment in accordance with a formula set out in legislation, through to highly specialised systems which use automated tools for prediction and deliberation, including the use of machine learning. There is no question that ADM has moved contemporary living into a new era which includes the evolution and move towards self-driving cars and all manner of ‘smart technology’ as well as an upsurge in the use of biometric data in some countries.[20]

ADM systems can be used in different ways in decision-making. For example, they can decide, recommend a decision to the decision-maker, or guide a decision-maker through relevant facts, legislation, and policy, thereby closing irrelevant paths through the process. This may include guidance and commentary relevant across different points of the decision-making process, enabling preliminary assessments for individuals or internal decision-makers and automation of some of the fact-finding processes which may influence subsequent decisions. For example, data from other sources or data that has been directly entered or uploaded to the system can be applied at key points of the process.[21]

C Application of Automated Decision-Making

However, in this article, ADM means when technology is used to make a firm and definitive decision, rather than a recommendation or part decision in which the final decision is left to an individual.[22] Here, ADM adopts a formal, legal, and binding decision-making aspect within the confines of power of some agency over the people it has jurisdiction over. As such, the scope and boundary specifications are narrow and are distinguished from other technology and data systems, encompassing a plethora of 'decisions' which an array of technology implement, including the decision that a patient's condition is lying outside of safe physiological parameters and needs a nurse’s attention, or that a dam gate needs to be raised or lowered to manage an impending flood based on the input of river height sensors, or which business solution to offer when managing investments and financial matters via Enterprise Resource Planning in the commercial banking industry.

Likewise, ADM extends from the use of simple rules-based formula to affirm when individuals meet objective criteria such as in employment screening or profiling processes, to using predictive algorithms, which encompass techniques including rules, but where a computer learns a model to decide through machine learning capabilities, rather than being programmed to execute a decision-making process in a specified way.

II WHY AND HOW DOES ROBODEBT FALL SHORT IN MAINTAINING ACCOUNTABLE AND RESPONSIBLE GOVERNING?

Despite some administrative and technical refinement of the OCI program, then rebadging the program, the core problematic issue remained. The burden of the onus of proof shifted onto individuals identified as having likely debts, based on the use of averaging to calculate overpayments and debts when those very individuals were often unable to provide evidence their original reporting was accurate.[23] The administrative burden this involved fell to groups of people with the least capacity to fulfil what was needed. Public outcry about this system meant the issue soon made its way into the public arena. “Noise” began within six months of OCI’s introduction, commencing with over 100 complaints to the Commonwealth Ombudsman by members of the public about letters they received from the DHS stating that they owed significant debts to the Government for past income support payments received.

The approach taken of continuing to pursue vulnerable citizens meant that the Government delegitimised itself, and the expected level of integrity and fairness required by those in public office, thereby undermining responsible government, and tainting democracy. Such would be the impact on Australian citizens that their negative sentiment and loss of faith in the Government’s treatment of people would be appropriated to how the public interacts with the whole of government across different organisations.[24] Initially, the then Human Services Minister Alan Tudge insisted the automated process was not flawed, claiming in excess of $AUD300 million worth of overpayments was correctly recovered, stating, “The system is working and we will continue with that system,” after being pursued by the opposition some fifty days earlier.[25]

Controversy about the Robodebt notices sent out mounted to the extent that the Auditor General was requested to investigate, and in 2017 an independent MP asked the Office of the Commonwealth Ombudsman (‘Ombudsman’) (the independent watchdog with authority to investigate the departments overseeing Robodebt) to launch an investigation. By January 2017, social media was flooded by outraged members of the public with accounts set up to contest the debts with individuals and journalists airing the issue on social and mainstream media. This prompted no government response,[26] indicating the need for a much more supportive culture of public opinion for the vulnerable.[27] The Commission heard from Ombudsman’s representatives, in 2017 about information that was withheld from the Ombudsman.

One theme that emerges about the failings of Robodebt is omission of information, and it was not until almost six years later at the commission that it was made clear key documents (2014 emails identifying legal issues about income averaging and indicating 76% of debts were calculated using a person’s average earnings) which would have closed the Robodebt scheme, were not provided to senior assistant ombudsman Louise Macleod in her Robodebt investigation. This failure by the DHS to provide essential information to the Ombudsman’s inquiry is a breach of legislation,[28] and a flagrant disregard for the authority of the Ombudsman's office which could have changed the outcome of the inquiry.[29]

The DHS also omitted words contained in a cabinet briefing which would have alerted ministers about the need for legislation to give Robodebt legitimacy. What is now known from the Commission hearings is that former prime minister Malcolm Turnbull was concerned about the accuracy of information and the degree of fairness which resulted in cases where it was hard to prove the facts, as supported by his 7 January 2017 comments to Alan Tudge. Despite pressing Mr Tudge on the method used, it seems the Government was more concerned about keeping the ability level of the public service intact, rather than calling in external consultants to do what Malcolm Turnbull considered as their work and undermining their abilities. However, because Robodebt was presented to cabinet under the guise of legislation it was not questioned as Mr Turnbull stated, “I did not turn my mind to the legality of the program,[30] [as] ... AGS has advised on the legality of the Scheme and no legislation is required”.[31]

Subsequently, a Senate Community Affairs Reference Committee Inquiry, reported that between November 2016 and March 2017 at least 200,000 people were affected, with 20,000 debt letters being sent out per week. This meant a quadrupling of ‘debt interventions’ compared to the previous full year and almost twice the amount of debts raised comparing 2014-2015 to 2015-2016. The Senate Committee report confirmed the shift in responsibility to the debt recipients for clarifying and checking their income information as one of three key changes involved in the OCI. The recipients were directed to an online portal to check the information and provide supporting evidence of their fortnightly income, retrospective to 2010 for some people.

The significant reduction in workload for the department by outsourcing of work confirmed a substantial workload reduction for staff enabling a massive increase in income discrepancy investigations.[32] The process of shifting the onus of proof to the alleged debtor is unlawful because Centrelink assumes the legal onus of ‘proving’ the existence and size of any debt denied by the alleged debtor.[33] Further, generally people in this group represent vulnerable and potentially marginalised groups in society who may not only be ill-equipped to grapple with what was required of them to support non-payment and erroneous debts to be paid but through the process were further marginalised and isolated. Unequivocally, the potential benefits of ADM as a powerful tool for dealing with large data sets enabling efficiency of scale, for cost effectiveness, precision, consistency, accuracy, and timeliness in the Government’s decision-making had not paid off financially. Also lacking were plans to mitigate and respond to risks in a concerted manner.[34]

The risks may have not been realised were the application used aptly for making entitlements and decisions reliant upon logic and rules-based programs that apply rigid criteria to designated varied circumstances and scenarios. In the planning phase, ADM benefits needed to be weighed up against threats including its legality questions and issues involving human cost and detriment. Clearly, this just did not happen because if it did it is inconceivable that things would have proceeded as they did. The human costs of dealing with debt notices by the recipients include emotional distress and undermining peoples’ health and wellbeing, and resource inadequacy as well as loss of anticipated efficiencies within the Government. Regulations for ADM need to ensure their compatibility with the central administrative law values and principles underpinning a democratic society governed by the ROL. These include how fit for purpose law and regulation is for making human centric decisions and protecting individual rights.[35]

In June 2021, a Federal Court Judge approved a settlement worth AUD$1.8 billion for people wrongly pursued by Robodebt. The court found the Commonwealth had unlawfully raised AUD$1.73 billion in debts against 433,000 people. Of this, $AUD751 million was wrongly recovered from 381,000 people. Settlement payments to eligible group members involved the Robodebt class action.[36] Serious questions are being asked about how the situation reached the scale it did and why things failed along the way. There have been calls for an apology, due to the complete lack of transparency that can be traced from the Prime Minister’s Office down, stemming from Scott Morrison’s secret self-appointments to multiple Ministerial positions without informing Parliament nor the public which have been criticised as undermining responsible Government and the entire democratic system of Government.[37]

High ranking public servants “should have known”[38] and now state they were aware of Robodebt’s unethical and potentially illegal status but chose inaction or “played down” the seriousness of the situation. The Royal Commission has unearthed evidence Kathryn Campbell Secretary of the Department of Social Security (‘DSS’) referred to Robodebt’s illegality as “legally insufficient” even after it ceased to operate. Her predecessor Serena Wilson conceded she “took no steps” to correct the situation, when armed with damming legal advice as early as 2014 about the legal status of Robodebt, while Secretary Finn Pratt, her manager, claimed the advice did not reach him.

However, there is no denying that Campbell knew from the end of 2014, in her role of administering the social security policy, that the OCI proposal that became Robodebt, required legislative amendment to be legal.[39] Another still unanswered question is how in March 2015 a version of the policy proposal for the OCI was sent to lawyers at DHS and it was recorded that legislative change was not required, raising warning bells with officials. It was then to be another two years until the story became big in the media and officials claimed they thought the income average method had been modified. [40]

Australian citizens place their trust in the Government, but as demonstrated by the notion of “hiding one’s head in the sand”, by ignoring important legal matters, and inaction including not passing on information is far from acting responsibly. Rather, an arrogant and reckless approach was adopted by public officials and left unchecked as the Robodebt inquiry heard "definitive" legal advice showing the OCI needed to change before proceeding but it did not.[41] In rolling out the OCI, officials never explicitly questioned the advice of income averaging. Income averaging, the method used to calculate debts, is unlawful. Moreover, the timespan over which things continued to operate unchecked was truly extraordinary.[42] The financial windfall coupled with the government’s need to “save face” kept it surging on in the same direction. However, had the government taken more time over the design and followed the guidelines set out by the then Administrative Review Council in its 2004 report on machine learning, it is difficult to imagine that such legal errors would have been made or persisted over time.[43]

Egregious in its failure, there appears to be no one culprit at the centre of the Robodebt debacle, but rather, the range of players and agencies involved who have been exposed. At this point, the public is wondering who should shoulder the responsibility. There is a degree of collusion hidden behind the veneer of convenience with the Ministers who oversaw Robodebt "... [circling] the same themes in answering questions at the Commission”. Typically, the response 'I did not know. I relied on the public servants to tell me about any risks they perceived', is consistently used as a line to defend Minister’s failure to stop Robodebt. Yet, it is true that "Despite having all the warnings, no written briefs were raised to ministers directly warning them”.[44]

III LESSONS LEARNED THROUGH THE ROYAL COMMISSION ABOUT USING ADM

In this section, three reform areas are analysed including how to address issues of bias, design and development of ADM and law, and policy reform. The rapid expansion in the use of ADM systems by governments[45] continues to raise novel legal questions and the questions surrounding Robodebt have brought these issues to the surface.[46] Therefore, it offers a potentially positive way to recognise the need for changes. It is time to strike a better balance between the requirements of efficiency, potential revenue streams and those of legality and justice to the individual, as ADM is best used when it reflects the complexity of the human experience,[47] including by implementing measures to protect vulnerable people within the welfare system rather than dismissing them. This balance encompasses not compromising on program design and standards, pursuing legal reform,[48] and adopting safeguards against the failure of due diligence by monitoring bodies including the Ombudsman. Such an approach includes testing and refining any new system, particularly one involving so many people to ensure technically and legally it goes through all the required checks and balances, rather than expeditiously rushing it through for political point scoring.

Likewise, structural change is needed to Centrelink’s functioning for it to operate more deliberately and meet its stated priorities of “... building on our foundation of service excellence to shape how government services are delivered to the Australian community”[49] and “providing payments and services that match customers’ circumstances”.[50] These priorities sit in sharp contrast to the human cost of Robodebt, as a recipient’s statement about receiving a DHS letter of debt and the consequences exemplifies, “I was literally crushed. I was in shock. I walked around my house trying to deny the reality of what had happened. ... I was so ashamed ... I didn't know who to talk to about this”. [51]

A Addressing Bias and Fairness

First, in using ADM to make administrative and binding decisions, the issues of bias, circumstantial information involving administrative decision-making and the adequacy of judicial review mechanisms to provide effective oversight of the use of such processes by the executive in decision-making warrants attention. The issue of power and human rights is vital for ADM and legislation is needed which includes a human rights impact assessment (‘HRIA’) as part of developing the legislation. An HRIA would increase accountability and responsible governing through public consultation, inclusive of groups of individuals to be affected by ADM. Likewise, HRIA should assess whether the proposed AI-informed decision-making system complies with Australia’s international human rights law obligations.[52]

It is essential legislation can ensure a right to merits review, before an independent tribunal such as the former Administrative Appeals Tribunal (soon to be introduced as a new federal administrative review body)[53], for any AI-informed administrative decision, given what has occurred in Robodebt. An accessible and known process to review and to resolve disputes is recommended. Additional safeguards for responsible government include introducing an expert body, such as the AI Safety Commissioner, to provide advice on good practice regarding human review, and monitoring of ADM. This body should also advise the Government on ways to promote high standards within Government.[54]

B Design and development of ADM

Second, are ADM design and development issues which need to be put through iterative processes of refinement, to test, pilot, gain feedback and refine what has been developed with adequate resourcing. This can involve failures and reworking or changing the way things are being developed which is not possible when ADM is developed quickly. In first constructing a beta model and adequately testing, technical flaws, bugs and errors could be alleviated in the future. This would overcome issues, such as accounting for different spelling of employer’s names across databases, the inability to rationalise a subjects’ work history, including casual work, and managing data from two incompatible databases.

These issues were further exacerbated by an unnecessarily complex and confusing interface regarding Robodebt.[55] Moreover, the Commission heard evidence from a witness who is a Deloitte representative regarding the firm's technical study of Robodebt and was told DHS had no risk management frameworks for the Robodebt program[56]. This seems like flying solo without a radio or checking the weather forecast first. Even more fundamental is the fact that the Commission heard further evidence from Deloitte that their review found AI was not present in the program. Rather, they concluded that AI was not evident in the program, and as it was a "relatively basic" automation system there was no capacity to learn from errors or improve accuracy over time.[57] The combined effect of these factors adds to the level of expediency and irresponsible conduct that are now regarded as hallmarks of Robodebt.

C Law and Policy

Third, involves the urgent need to amend law[58] and policy to ensure human decision-makers review original machine-made decisions, and make amendments to reduce the number of grounds on which departments can deny Freedom of Information (‘FOI’) requests concerning the use of algorithms in decision-making. The Government has faced several challenges involving the FOI legislation regarding the provision of access to the algorithms used in Robodebt. FOI laws permit individuals to request and gain access to government agency held documents, including algorithms.[59] However, the government has argued that the algorithms used in Robodebt are commercial-in-confidence and exempt from disclosure. The Act does require an agency to make information available on its operations by publishing information under the Information Publication Scheme and disclose a log of accessed information that has been accessed under FOI requests but there is sensitivity involving the algorithm.[60] However, the issue remains that ADM decisions are one step removed from individuals and therefore can threaten responsible governing, reducing the executive’s transparency and accountability which can harm the principle of responsible government.[61]

Additionally, a more proactive approach is needed in law than what currently already exists, to inform and advise the public of their rights and increase their knowledge and understanding of the issues involved.[62]

Amendments could clarify how the law operates with respect to automated systems and help to balance the objectives of the Act.[63] The public right to know also needs to be balanced against cybersecurity which is an ever-growing threat to the Government with data breaches continuing to increase across industries. The relationship between the FOI Act and automated systems is unclear because even though it seems that s10 may include automated systems, the operation of s10 regarding ADM systems currently requires operational information to be published which is problematic.[64]

IV CONCLUSION

The flawed Robodebt Scheme is undoubtedly a public policy fiasco,[65] and represents a “shameful chapter in the administration of the Commonwealth social security system [as well as] a massive failure of public administration”.[66] Although it is yet to be seen whether any charges will be handed down because of the Commission’s findings by the Office of the Director of Public Prosecutions and/or whether there will be a recommendation of an internal DHS review.

The Commission cost a significant amount of money and increased the lack of public trust of the Government, particularly by vulnerable groups. With more than forty two days of hearings at the Commission commencing with the initial hearing on 27 September 2022 through to the fourth hearings concluding on 10 March 2023 it has been exhaustive, expensive, and difficult due to the time span involved, the magnitude of the issues and the number of witnesses involved in the hearings, yet necessary to protect people and ensure and ADM and AI is used for a proper purpose and within the law.[67] After Attorney-General Mark Dreyfus approved the expenses to be paid for by the taxpayer for six former Coalition frontbenchers which also involved two former prime ministers, public servants, and finance and technology experts,[68] it is imperative that changes come from the proceedings of the Commission inquiry to help improve how the public service and the Government operate in future.

Robodebt provides an example of the harm that results when safeguards protecting the ROL fail and Government expediency, arrogance and a public policy with an inherently unethical character is adopted. Moreover, it demonstrates how when risks are not factored in or plans put in place to meet challenges that arise, there can be harmful outcomes which outweigh the benefits of ADM. While ADM’s differentiation point is evident in its responsiveness to predetermined outcomes married against factual scenarios resulting from user input of entered information, in contrast, administrative decisions require the exercise of a discretion or evaluative judgment and are a poor fit for ADM – as such Robodebt was completely flawed.

Achieving efficiency by using ADM has the potential to compromise key legal principles such as a commitment to an administrative law system that is fair, transparent, and accountable and operates according to the ROL. Safeguards need to be in place, to ensure that the individual’s right to hold decision makers to their obligations is preserved. The Government needs to be held to account to ensure the legality of purported actions by public bodies, to guard against the undermining of procedural fairness.[69] Safeguards, transparency, and accountability of government decisions by the provision of reasons and effective access to merits and judicial review are central tenets of how a government should conduct itself, yet recent examples including Robodebt support the need for review and the development of more robust regulation.[70]

It is recommended that future research continues into ways to empower and strengthen the voice of vulnerable individuals in the welfare group, as often they have no voice and examines how to increase internal accountability government mechanisms as well as good practice principles for developing ADM systems in the Government.


[1] Terry Carney, ‘Robo-debt Illegality: The Seven Veils of Failed Guarantees of the Rule of Law’ (2019) 44(1) Alternative Law Journal 4.

[2] James Kerr, Pintarich v Deputy Commissioner of Taxation [2018] FCAFC 79; (2018) 262 FCR 41 [47].

[3] Luke Henriques-Gomes, ‘Robodebt Likely a ‘Stuff Up’ Rather Than a ‘Conspiracy’, Former Department Boss Tells Inquiry’ (2022) The Guardian (online, 10 November 2022).

[4] Special Broadcasting Service (‘SBS’), ‘Legal Advice Showed the Robodebt Scheme Needed Changes to Proceed. An Inquiry has Heard It Wasn't Passed On’ (2022) SBS Australia News (online, 3 November 2022) <https://www.sbs.com.au/news/article/legal-advice-showed-the-robodebt-scheme-needed-changes-to-proceed-an-inquiry-has-heard-it-wasnt-passed-on/ibepdjyu3> (‘SBS Australia News’).

[5] Australian Government Royal Commission into the Robodebt Scheme (Web Page, 2023) <https://robodebt.royalcommission.gov.au/hearings>.

[6] Valerie Braithwaite, ‘Beyond the Bubble That is Robodebt: How Governments That Lose Integrity Threaten Democracy’ (2020) Australian Journal of Social Issues 242.

[7] Luke Henrique-Gomes, in Nikidehaghani, Mona, Andrew, Jane and Cortese, Corine, ‘Algorithmic Accountability: Robodebt and the Making of Welfare Cheats’ (2022) 36(2)Accounting, Auditing & Accountability Journal 677.

[8] Gareth, Hutchens, ‘The Phrase ‘Dole Bludger’ Emerged in the 1970s, and It’s Still Serving Its Political Purpose’ Australian Broadcasting Corporation (online, 30 May 2021) <https://www.abc.net.au/news/2021-05-30/dole-bludger-emerged-in-the-1970s-to-serve-a-political-purpose/100174356 >.

[9] Luke Henriques-Gomes, ‘Conspiracy, Or Stuff Up? Robodebt Royal Commission Probes How Far Up The Chain of Command Blame Falls’ (2022) The Guardian (online, 12 November 2022) <https://www.theguardian.com/australia-news/2022/nov/12/conspiracy-or-stuff-up-robodebt-royal-commission-probes-how-far-up-the-chain-of-command-blame-falls >.

[10] Braithwaite (n 6).

[11] Tom Tyler, Why People Obey the Law Princeton University Press, 2006.

[12] Tapani Rinta-Kahila et al, ‘How to Avoid Algorithmic Decision-making Mistakes: Lessons from the Robodebt’ Debacle’ (2022) 31(3) European Journal of Information Systems 317.

[13] Peter Whiteford, ‘Debt by Design: The Anatomy of a Social Policy Fiasco – Or was It Something Worse?’ (2021) 80 Australian Journal of Public Administration 341.

[14] Ibid 342.

[15] Social Security (Administration) Act 1999 (Cth).

[16] Whiteford (n 13) 340.

[17] Gordon Legal in Peter, Whiteford, ‘Debt by Design: The Anatomy of a Social Policy Fiasco – Or Was It Something Worse?’ (2021) 80 Australian Journal of Public Administration 80, 340.

[18] Australian Government Royal Commission into the Robodebt Scheme, ‘Extension of the Robodebt Royal Commission’ (Web Page, 2023) <https://ministers.ag.gov.au/media-centre/extension-robodebt-royal-commission-16-02-2023 >.

[19] Commonwealth Ombudsman, Automated Decision-Making - Better Practice (2007) Guide <https://www.ombudsman.gov.au/__data/assets/pdf_file/0029/288236/OMB1188-Automated-Decision-Making-Report_Final-A1898885.pdf> (‘Commonwealth Ombudsman’).

[20] Jack Stilgoe, ‘Machine Learning, Social Learning and the Governance of Self-driving Cars’ (2017) 48(1) Social Studies of Science 25.

[21] Commonwealth Ombudsman (n 19)10.

[22] Australian Government, Positioning Australia as a Leader in Digital Economy Regulation (Issues Paper, 2022) Commonwealth of Australia 3.

[23] Whiteford (n 13) 345.

[24] Braithwaite (n 6).

[25] The Courier Mail, ‘Human Services Minister Alan Tudge Defends Controversial Centrelink System’ (online, January 11 2017) <https://www.couriermail.com.au/news/human-services-minister-alan-tudge-defends-controversial-centrelink-system/news-story/70a22373c0c11afada384376c52691fe>.

[26] Whiteford (n 13) 345.

[27] Carney (n 1) 8.

[28] Public Service Act 1999 s 10(1)a, d–f.

[29] Maeve Bannister, ‘Final Week of Robodebt Royal Commission Public Hearings’ (online, March 8 2023) <https://www.perthnow.com.au/politics/final-week-of-robodebt-royal-commission-public-hearings-c-9979244 >.

[30] Australian Government, (2023) 20–30, 4426.

[31] Ibid.

[32] Senate Community Affairs Reference Committee, Design, Scope, Cost-benefit Analysis, Contracts Awarded, and Implementation Associated with the Better Management of the Social Welfare System Initiative (2017) 17 Commonwealth of Australia 2.21.

[33] Terry Carney, ‘The New Digital Future for Welfare: Debts without Legal Proof or Moral Authority?’ (2018) University of New South Wales Law Journal Forum 1, 4 <https://www.unswlawjournal.unsw.edu.au/wp-content/uploads/2018/12/2018-1-CARNEY.pdf >.

[34] Anna, Huggins, ‘Addressing Disconnection: Automated Decision-Making, Administrative Law and Regulatory Reform’ University of New South Wales Law Journal (2020) 43 (3) 1048, Melissa, Perry and Sonya, Campbell, ‘AI and Automated Decision-Making: Are You Just Another Number? Kerr’s Vision Splendid for Administrative Law: Still Fit for Purpose?’ Gilbert & Tobin Centre of Public Law, UNSW Law & Justice & NSW Chapter, Australian Institute of Administrative Law Friday, 21 October 2021.

[35] Ibid 1048.

[36] Australian Government Class Action Settlement, (2022) <https://www.servicesaustralia.gov.au/information-for-people-who-got-class-action-settlement-notice?context=60271>.

[37] Australian Broadcasting Commission, ‘Scott Morrison's Secret Ministries: What We Learned From the Solicitor-General's Advice’(2022) ABC News (online, 23 August 2022) <https://www.abc.net.au/news/2022-08-23/scott-morrison-secret-ministries-solicitor-general-investigation/101360028>.

[38] Prygodicz v Commonwealth of Australia (No 2) [2021] FCA 634 [6] (Murphy J).

[39] Henriques-Gomes (n 9).

[40] Ibid.

[41] SBS Australia News (n 4).

[42] Carney (n 1) 5.

[43] Ibid

[44] Darren O’Donovan, in Alexander Lewis 2023 ‘What We've Learnt From Nine Weeks of Robodebt Royal Commission Hearings,’ (2023) ABC News (online, March 11 2023) <https://www.abc.net.au/news/2023-03-11/robodebt-scheme-government-royal-commission-fraud/102074840>.

[45] Lyndal Naomi, Sleep, ‘From Making Automated Decision Making Visible to Mapping the Unknowable Human: Counter-Mapping Automated Decision Making in Social Services in Australia’ (2022) 28(7) Qualitative Inquiry 848.

[46] Will Bateman, ‘Algorithmic Decision-Making and Legality: Public Law Dimensions’ (2020) 94(7) Australian Law Journal 520–530; Joe, Ludwig, ‘The Freedom of Information Act: No Longer A Substantial Disappointment’ Admin Review (2010) 59(4) 5‑–18; Alexander, Jonathan, Brown, ‘Public Interest Disclosure Legislation in Australia: Towards the Next Generation An Issues Paper’ Commonwealth Ombudsman (2006) 1–58; Carney (n 1) 4–10.

[47] Melissa Perry and Sonya Campbell, ‘AI and Automated Decision-Making: Are You Just Another Number? Kerr’s Vision Splendid for Administrative Law: Still Fit for Purpose?’ Gilbert & Tobin Centre of Public Law, UNSW Law & Justice & NSW Chapter, Australian Institute of Administrative Law Friday, 21 October 2021.

[48] Andrew Ray, ‘Implications of the Future Use of Machine Learning in Complex Government Decision-making in Australia’ (2020) 1(1) Australian National University Journal of Law and Technology 1.

[49] Australian Government Services Australia (Web Page, 2023) <https://www.servicesaustralia.gov.au>.

[50] Ibid.

[51] Name withheld, Submission 50, 1 The Senate ‘Accountability and Justice: Why We need a Royal Commission into Robodebt Commonwealth of Australia’ (2022).

[52] Australian Human Rights Commission, AI-informed Decision-making (Web Page, 2023) <https://tech.humanrights.gov.au/artificial-intelligence/ai-informed-decision-making>.

[53] Australian Government Administrative Appeals Tribunal (Web Page, 2023) <https://www.aat.gov.au/>.

[54] Ibid.

[55] University of Queensland, ‘How to Avoid algorithmic Decision-making Mistakes: Lessons from Robodebt’ (online, 28 April 2022) <https://bel.uq.edu.au/article/2022/04/how-avoid-algorithmic-decision-making-mistakes-lessons-robodebt-debacle >.

[56] Bannister (n 29).

[57] Ibid.

[58] Freedom of Information Act 1982 (Cth) (‘FOI’).

[59] Ibid.

[60] Australian Government, The Treasury (Web Page, 2023) <https://treasury.gov.au/the-department/accountability-reporting/foi>.

[61] Ray (n 48).

[62] Government Information (Public Access) Act 2009 (NSW) (‘GIPA’).

[63] FOI (n 58).

[64] Ray (n 48).

[65] Commonwealth Ombudsman, ‘Centrelink’s Automated Debt Raising and Recovery System’ (Report No 2, April 2017) <https://www.ombudsman.gov.au/__data/assets/pdf_file/0022/43528/Report-Centrelinks-automated-debt-raising-and-recovery-system-April-2017.pdf>.

[66] Prygodicz v Commonwealth of Australia (No 2) [2021] FCA 634 [5] (Murphy J).

[67] Australian Government Royal Commission into the Robodebt Scheme (Web Page, 2023) <https://robodebt.royalcommission.gov.au/hearings>.

[68] Paul Karp and Luke Henrique-Gomes, 2022 ‘Taxpayers to Fund Legal Costs of Scott Morrison and other Former Ministers Related to Robodebt Royal Commission’ (The Guardian, Wednesday 23 November 2022) <https://amp.theguardian.com/australia-news/2022/nov/23/taxpayers-to-fund-legal-costs-of-scott-morrison-and-other-former-ministers-related-to-robodebt-royal-commission>.

[69] Ibid 1049.

[70] Ray (n 48)12.


AustLII: Copyright Policy | Disclaimers | Privacy Policy | Feedback
URL: http://www.austlii.edu.au/au/journals/UNSWLawJlStuS/2023/6.html