Automated Decision Making: Impacts on Principles of Administrative Law

Showing that something can be done does not mean that it should be done.[1]

Author: Vanessa Mihov, LLB (Hons), BBus (Econ)

You can visit Vanessa’s Linkedin here.

Introduction

Rapid expansions in information technology have transformed the legal system. Complex decisions being made without any requirement of human mental processes are quickly becoming the ‘new normal’.[2] This automated decision-making (‘ADM’) is achieved partly, or wholly, through computers that manipulate pre-set logical parameters to perform actions or make decisions, without the involvement of a human.[3] A growing number of statutes explicitly provide for computer programs to make decisions that stand as the decision of an agency of government.[4]

Whilst the expansion of information technology undoubtedly results in time and cost savings, ADM undermines the purpose of administrative law – to provide good government according to law – threatening its underlying ideals of fairness, accountability, and impartiality upheld by the rule of law and the doctrine of separation of powers.[5] Focusing on these fundamental administrative principles, this essay will analyse the potential impacts of the use of information technology, particularly in the form of ADM, in Australian public law. It will argue that ADM increases the number of grounds for judicial review available to individuals under section 5 of the Administrative Decisions (Judicial Review) Act 1977 (‘the ADJR Act’) through its lack of transparency, fairness, and lawfulness. It will conclude that with respect to these principles, ADM threatens the purpose of administrative law and its implementation should be limited to ensure these values are not compromised.

Automated-Decision Making and Judicial Review

The design of ADM systems is such that if used without human regulation, its limitations give rise to judicial review on the grounds of lack of evidence, error, and procedural unfairness.[6] The right to judicial review is entrenched in the common law, the ADJR Act, the Judiciary Act 1903, and the Constitution.[7] The purpose of judicial review is to review an exercise of executive power to determine whether the decision-maker used the correct legal reasoning or followed the correct legal procedures. Upon successful review, the remedy is to quash the decision and order the decision be remade correctly and according to the law.[8] Pursuant to the separation of powers doctrine, the executive bears the onus to use the reasons from the judiciary to remake the decision correctly and potentially amend legislation where necessary.

This process ensures that the rights of applicants are protected in their dealings with government agencies and improves the quality of government decision-making.[9] However, the limitations of ADM systems threaten this administrative process by creating unjustified decisions that are unfair and unlawful.

Lack of transparency cause breaches of Natural Justice

ADM does not provide for transparency due to the absence of reasons provided to individuals when making decisions, violating the principles of the rule of law.[10]ADM systems cannot readily discern why any particular prediction resulted from artificial intelligence, therefore, causing difficulties in determining what evidence is needed to correct errors.[11] Common law cases have demonstrated that initial decision-makers are sometimes unable to provide explanations to support government decisions.[12] The impacts of this are twofold. Firstly, the separation of powers is undermined if the government is not held accountable for its decisions and actions by way of providing reasons. The absence of reasons due to ADM cannot guarantee that the government is acting lawfully and not exceeding its power, undermining the rule of law that prima facie, all people are considered equal before the law. Secondly, unless provided with a plain English explanation of the basis for the decision, a person aggrieved by a decision may not understand nor adequately challenge the decision available under the hearing aspect of natural justice.[13] This directly refutes the underlying administrative law principles of fairness and the judicial system providing procedural fairness through a fair and prompt trial.[14]

The Hearing Aspect of Natural Justice

Procedural fairness (or natural justice) forms the basis for a ground of judicial review under the common law and the ADJR Act, requiring an individual to receive a fair hearing.[15] In Amato v The Commonwealth, the Department of Human Services’ use of ADM system (‘robo-debt’) caused breaches of natural justice where inadequate testing of the website caused difficulties for diverse welfare recipients using the system and insufficient information meant individuals could not challenge the incorrect debts being raised against them.[16] The complete absence of evidence for a finding of fact is an error of law giving rise to another ground of judicial review reflected in s 5(1)(f) and under the first limb of s 5(1)(h) of the ADJR Act. A decision can be quashed under a writ of certiorari if there is complete absence of evidence to justify the decision.[17] However, increases in claims under either reason for review pose a risk of unnecessary pressure on the judiciary which can be avoided by not employing ADM systems that cannot produce reasons for a decision. This mirrors international standards where it is not acceptable to employ any artificial intelligence system which could have a substantial impact on an individual, unless it can generate a full and satisfactory explanation for the decision.[18] Ultimately, ADM undermines the administrative legal system causing decision-makers to breach legal principles to avoid responsibility and the quality of judicial review, serving as a dynamic feedback loop for lawmakers, to deteriorate.[19]

The Bias Aspect of Natural Justice

ADM systems have proven to generate biased decisions, compromising the administrative principle of fairness and giving rise to judicial review under the bias aspect of natural justice.[20]

ADM systems respond to inputs entered by a decision-maker in accordance with predetermined outcomes which raises the potential of encoding unfair, discriminatory decisions due to the social scoring system in place.[21] Furthermore, administrative decisions require the exercise of discretion which is beyond the capacity of an automated system whose predetermined outcomes may be characterised as pre-judgment or bias.[22] Even if the system is regulated by human power, it is possible for a decision-maker to input data deliberately to obtain an outcome adverse to the individual’s interests which questions the legitimacy of the decision.[23] This gives rise to the biased aspect of the natural justice ground of review threatening the administrative principle of fairness, particularly where power imbalances exist between the decision-maker and an aggrieved who may already be in positions of economic and/or social disadvantage.[24] Ultimately, ADM systems directly oppose natural justice as its procedures are unfair and inappropriate, causing further ethical issues and violating the rule of law that all persons are equal before the law.[25]

System errors raise issues of legality

ADM threatens the administrative law principle that government decisions must be lawful due to errors in translation of complex laws into code.[26] When the ADM process does not include human oversight, the decision is more prone to error.[27]The Australian Department of Veterans’ Affairs utilise ADM systems to apply thousands of pages of legislation and automate aspects of determining compensation claims.[28] This allows the Department to determine 30% more claims annually using 30% fewer human resources in substantially less time, saving approximately $6 million annually in costs.[29] However, these decisions are more prone to judicial review under error of law because the decision-maker (now the ADM system) can misinterpret statutory provisions, or misapply a principle of law.[30] In Amato v Commonwealth, the Court ruled that how the robo-debt system calculated Centrelink debts on the basis of averaging income data was inaccurate and unlawful.[31]

This case highlights a major concern regarding the legitimacy of decisions calculated by other government agencies employing ADM processes and whether these programming errors may be replicated among thousands of decisions undetected.[32] Such was the case in the United States where an error in the Benefits Management System caused hundreds of thousands of erroneously calculated Medicaid, welfare and benefits decisions.[33] Had the system failure been less serious the invalid decisions may have gone unnoticed, as they often do in Australia as individuals trust and assume that decisions are rightfully issued by the government agency.[34] This raises a significant issue of legality with decisions made through ADM systems and undermines the administrate law principle of trust in government, which is essential for effective and efficient policy making and a functioning society.[35] The impacts of this include risks to the operation of the rule of law, as well as lower rates of compliance with rules and regulations causing a chain reaction of social, economic, and political effects.[36]

Conflicts between the Judiciary and Executive

The errors in ADM systems have also proved issues in legality which threaten the doctrine of separation of powers. In Pintarich v Deputy Commissioner of Taxation, the Court ruled that it is unlawful for a decision-maker to renounce a decision that was sent to an individual through an ADM system.[37] Because of this case, under common law, a valid decision requires both a mental process of reaching a conclusion and an objective manifestation of that conclusion.[38] However, statutes including provisions for ADM often exclude common law impacting its operation. In the Australian Citizenship Act 1907, a Minister can substitute a decision made by an automated system with another decision which has effect despite any rule of common law to the contrary effect.[39] This causes a conflict between the legislature and Court because if legislation excludes administrative law principles the High Court has original jurisdiction under s 75(v) of the Constitution securing the basic element of the rule of law but undermining the separation of powers.[40]

Recommendation

Implementing ADM systems to make decisions on behalf of government agencies without human oversight increases the number of grounds for judicial review available to individuals. The Department of Human Services alone will undertake almost 40 times more compliance interventions per year under ADM processes than under the previous manual process.[41] These interventions may require legal advice which causes further impacts on access to justice for individuals who may not be able to secure legal aid that is inundated with cases for review. Such were the effects of Centrelink’s robo-debt system where calls for legal help to Victorian Legal Aid increased by 300 per cent.[42] However, the time and cost savings undoubtedly improve efficiency in decision-making processes which cannot be ignored.

Therefore, in order to reap the benefits of ADM without increasing grounds for judicial review and impacting transparency, fairness, and legality, it is vital to prevent the reasons for these grounds from arising in the first place. To avoid impacts on transparency and natural justice, systems that are not yet capable of providing explanations for their decisions should not be employed until this becomes available.[43] In terms of fairness, ADM systems should only be implemented for calculations and non-discretion-based decisions to save time and resources. Allowing an algorithm to determine outcomes requiring discretion defeats the purpose of administrative law and should only be exercised by humans. In order to safeguard decisions from error and the issue of legality in ADM processes, proper verification and regular auditing should be integrated into the systems as well as appropriate mechanisms for human review put in place.[44] To prevent conflicts between the separation of powers, statutes should be drafted to not exclude common law affecting its operation. With these precautions, ADM processes can achieve efficiencies in time and costs, as well as promote transparency and accountability to promote a comprehensive administrative system.[45]

Conclusion

The purpose of the administrative system is to provide fair, efficient and effective decision making which is ultimately barred by the use of ADM without human regulation. This essay has identified that a lack of transparency and fairness in ADM systems causes breaches of both the hearing and bias aspects of natural justice. It also highlighted that system errors raise issues of legality as well as undermine the separation of powers. Ultimately, ADM can promote these underlying administrative law principles with appropriate precautions in place to ensure government decisions are informative, fair, and lawful to enhance and develop the administrative legal system.

[1] Kenneth Hayne, ‘Change’ (2017) 42(2) Alternative Law Journal 87, 88. [2] Pintarich v Deputy Commissioner of Taxation [2018] FCAF 79, 47. [3] Commonwealth Ombudsman, Automated Decision-making – Better Practice Guide (Report No 46, February 2007), 5 (‘Commonwealth Better Practice Guide’). [4] Social Security (Administration) Act 1999 (Cth) s 6A; Migration Act 1958 (Cth) s 495A; Australian Citizenship Act 2007 (Cth) s 48. [5] Mark Aronson and Matthew Groves, Judicial Review of Administrative Action (Thomson Reuters, 5th ed, 2013) 1. [6] Administrative Decisions (Judicial Review) Act 1977 (Cth) ss 5(1)(a), 5(1)(f), 5(1)(h) ‘(ADJR Act’). [7] ADJR Act (n 6) s 5; Judiciary Act 1903 (Cth) s 39B; Commonwealth of Australia Constitution Act (Cth) ss 75(v), 75(iii). [8] ADJR Act (n 6) s 16. [9] Administrative Review Council, Automated Assistance in Administrative Decision Making (Report No 46, November 2004) 8 (‘Administrative Review Council report’). [10] Melissa Perry ‘iDecide: The Legal Implications of Automated Decision-making’ (Speech, University of Cambridge, 15 September 2014), 3 (‘iDecide Speech’). [11] Michèle Finck, Automated Decision-Making and Administrative Law, (Research Paper No. 19-10, 5 August 2019) 13 (‘Finck report’). [12] Re Schouten and Secretary, Department of Education, Employment and Workplace Relations [2011] AATA 365, 39. [13] Re Schouten (n 11) 39; ADJR Act (n 6) s 5(1)(a). [14] Gabrielle Appleby et al, ‘Contemporary Challenges Facing the Australian Judiciary: An Empirical Interruption’ (2019) 42(2) Melbourne University Law Review 2, 4. [15] ADJR Act (n 6) s 5(1)(a); Judiciary Act 1903 (Cth) s 39B; Commonwealth of Australia Constitution Act (Cth) ss 75(v), 75(iii). [16] Amato v Commonwealth VID611/2019; Parliament of Australia, ‘Design, scope, cost-benefit analysis, contracts awarded and implementation associated with the Better Management of the Social Welfare System initiative’, Senate Standing Committee on Community Affairs (Web Page, 8 February 2017) <https://www.aph.gov.au/Parliamentary_Business/Committees/Senate/Community_Affairs/SocialWelfareSystem>. [17] R v Australian Stevedoring Industry Board (1953) 88 CLR 100; Australian Broadcasting Tribunal v Bond (1990) 170 CLR 321, 95. [18] House of Lords, AI in the UK: ready, willing and able? (Report of Session 2017-19, 16 April 2018) [105]. [19] Finck report (n 10) 12. [20] ADJR Act (n 6) s 5(1)(a); International Finance Trust Co Ltd v New South Wales Crime Commission (2009) 240 CLR 319, 354 [54]. [21] Finck report (n 10) 11; iDecide Speech (n 9) 7. [22] iDecide Speech (n 9) 7. [23] Administrative Review Council report (n 8) 24. [24] Law Council of Australia, Submission 60 to Senate Standing Committee on Community Affairs, Centrelink’s Compliance Program (31 October 2019) [88]. [25] Kioa v West (1985) 150 CLR 550, 585. [26] iDecide Speech (n 9) 3. [27]Law Council of Australia, Submission 60 to Senate Standing Committee on Community Affairs, Centrelink’s Compliance Program (31 October 2019) [88]. [28] Department of Human Services, Annual Report 2012-13 (Final Report, July 2013) 22. [29] iDecide Speech (n 9) 6. [30] ADJR Act (n 6) s 5(1)(f). [31] Parliament of Australia, ‘Design, scope, cost-benefit analysis, contracts awarded and implementation associated with the Better Management of the Social Welfare System initiative’, Senate Standing Committee on Community Affairs (Web Page, 8 February 2017) <https://www.aph.gov.au/Parliamentary_Business/Committees/Senate/Community_Affairs/SocialWelfareSystem>. [32] iDecide Speech (n 9) 6. [33] Ibid. [34] Amato v Commonwealth VID611/2019. [35] OECD, Government at a Glance 2013 (Final Report, 10 November 2013) 20. [36] Ibid. [37] Pintarich v Deputy Commissioner of Taxation [2018] FCAF 79, 47. [38] Ibid 140. [39] Australian Citizenship Act 2007 (Cth) ss 48(3), 48(5). [40] Commonwealth of Australia Constitution Act (Cth) ss 75(v). [41] Kate Galloway, ‘Big Data: A Case Study of Disruption and Government Power’ (2017) 42 Alternative Law Journal 89, 93–4. [42] Miles Browne and Charley Brumby-Rendell, ‘An in-depth look at our robo-debt test case,’ Victoria Legal Aid (Web Page, 14 April 2020) . [43] Re Schouten and Secretary, Department of Education, Employment and Workplace Relations [2011] AATA 365, 39. [44] iDecide Speech (n 9) 6. [45] Ibid.