TLDR: This research paper explores the vital role of public interest litigation in promoting accountability for AI and automated decision-making (ADM) in Australia. It details how existing laws are being ‘retrofitted’ to address new technological harms, using examples like the Robodebt class action. Based on interviews with legal experts and activists, the study outlines pragmatic strategies and tactics for effective litigation, while also confronting its limits, such as problematic discrimination laws and the heavy burdens on litigants. The paper concludes by highlighting essential enablers for a thriving accountability ecosystem: transparency, robust information-sharing networks, and crucial funding support, advocating for urgent regulatory reforms alongside legal action.
In an era where automated decision-making (ADM) and artificial intelligence (AI) systems are increasingly influencing critical aspects of our lives, from welfare benefits to financial services, ensuring accountability for their potential harms has become a pressing global challenge. A recent research paper, titled “Retrofitters, pragmatists and activists Public interest litigation for accountable automated decision -making,” delves into the crucial role of public interest litigation in Australia in holding these powerful systems to account. Authored by Henry Fraser and Zahra Stardust, this paper highlights how existing legal frameworks are being adapted to address the novel injustices brought about by AI and ADM, especially when dedicated regulations lag behind.
The paper positions public interest litigation not as a standalone solution, but as a vital component within a broader ecosystem striving for transparency, accountability, and justice in the realm of ADM. It draws on insights from interviews with Australian public interest litigators, technology policy activists, and technology law scholars, offering a pragmatic view on how legal action can drive change.
The Challenge of Automated Harms
The research points to real-world examples of ADM systems, such as SyRI, MiDAS, and Australia’s own Robodebt, which were designed to detect welfare overpayments but ended up harming tens of thousands of economically vulnerable individuals. These systems were often implemented with a disregard for existing laws and a severe lack of empathy. While administrative reviews and reports failed to halt these injustices, litigation proved to be a powerful catalyst for change. The Robodebt class action, for instance, resulted in a significant settlement, with the government agreeing to repay over $751 million and withdraw over $1 billion in improper debt notices.
Beyond welfare, the paper notes numerous other serious harms emerging from ADM misuse over the past five years, including issues in law enforcement, autonomous vehicle accidents, toxic AI chatbots, widespread facial recognition surveillance, denial of financial services, and discriminatory decision-making in education and finance. The sheer scale and speed at which these tools process data and make life-changing decisions underscore the urgent need for effective accountability.
Retrofitting Old Laws for New Technologies
A central theme of the paper is the concept of “legal retrofitting” – adapting existing laws, some conceived centuries ago, to address contemporary problems posed by automated systems. Australia, lacking a federal Bill or Charter of Rights, often requires litigants to creatively apply constitutional law, common law, or older legislation to new circumstances. For example, property law was used to challenge the arbitrary confiscation of phones in immigration detention, demonstrating how lawyers must find innovative ways to use available legal tools.
This pragmatic approach involves selecting causes of action that lend themselves well to existing law, considering the type of harm to be remedied, and the available remedies. Doctrines like negligence, unjust enrichment, and breach of confidence are often favored because they can provide monetary damages, which are crucial for incentivizing litigation funders and law firms. The paper also discusses “bolt-ons,” where broader public interest matters or creative, untested legal claims are added to private disputes, diversifying the prospects of success and creating precedents for future cases.
Strategies and Limits of Public Interest Litigation
Participants in the study identified several goals for public interest litigation, including asserting that existing laws apply to new technologies, developing the law through new interpretations, and advocating for policy and legislative changes. Litigation, especially when integrated with broader advocacy and activism, can galvanize public opinion and demonstrate when existing laws are insufficient, paving the way for reform.
However, the paper also acknowledges significant limits. Australian anti-discrimination law, for instance, struggles to address algorithmic bias because it often focuses on individual intent rather than systemic discrimination, and typically doesn’t yield large monetary awards. The opacity of AI systems and a phenomenon called “technology deference” (the assumption that computers are less biased than humans) further complicate legal challenges. The legal culture itself, with varying risk tolerances among lawyers and judges, also plays a role.
Access to justice is another major barrier. Litigation is costly, time-consuming, and emotionally draining, especially for individuals already experiencing marginalization. The risk of adverse costs orders can deter potential litigants, particularly lead plaintiffs in class actions who bear sole liability. Furthermore, the immediate interests of a litigant (e.g., a quick settlement) can sometimes conflict with the broader public interest in establishing legal precedents or exposing systemic issues.
Also Read:
- AI Innovation and Human Rights: A Call for Responsible Governance
- Navigating AI Content: How Europe’s AI Act Shapes LLM Watermarking
Enabling a Robust Accountability Ecosystem
To overcome these challenges, the paper identifies three essential enablers for effective public interest litigation:
1. Transparency Throughout the Accountability Lifecycle: Stakeholders need various kinds of transparency at multiple points, from basic notice of ADM use to detailed explanations of how systems work. This includes documentation, impact assessments, and inspectability of algorithms, which can create a “paper trail” for enforcement. Transparency also facilitates “peer review” by researchers, regulators, and journalists, raising community awareness and helping to detect patterns of wrongdoing.
2. Networks for Information Sharing and Community Mobilisation: The fragmented nature of complaints and regulatory silos means systemic issues often go unnoticed. There’s an urgent need for mechanisms to receive, aggregate, and analyze complaints to detect patterns of harm. Formal institutions for systematic monitoring and review, like Australia’s reconstituted Administrative Review Council, are crucial. Equally important are informal networks, “hackathons,” and specialist digital legal centers that connect affected individuals, community organizations, and lawyers to share information, develop legal theories, and mobilize stakeholders.
3. Funding Support for Litigation and Litigants: The prohibitive costs of litigation necessitate better funding for civil society organizations, community legal centers, and Legal Aid. The paper suggests exploring options like a universal public legal services system. It also advocates for reforms to the adverse costs regime, such as public interest exceptions or protective costs orders (similar to those in the UK), to reduce the financial burden on litigants and encourage justice-seeking actions. Beyond financial support, a robust social infrastructure is needed to assist litigants with the personal strain of legal battles.
The paper concludes by emphasizing that while public interest litigation is often a last resort, it is also a generative force, exposing data, building momentum, and planting seeds for broader social movements. It underscores the urgent need for comprehensive law reform, including “mandatory guardrails” for high-risk AI applications, to prevent harm from automated decision-making and ensure these systems work in service of marginalized communities and social justice. For more details, you can read the full research paper here.
