Automated Decision-Making and Migration: Can ADM Engineer Fairer Outcomes?

Amrita Chahal | Cyber, Tech and Space Fellow

Image sourced from Pascal Meier via Unsplash 

As automated decision-making (ADM) has captured the attention of the international community, governments and industries alike are looking to harness its power to increase productivity. The appeal is understandable for public administrators, many of whom manage large volumes of casework and critical deadlines. For the Department of Home Affairs, that casework includes thousands of new humanitarian applicants each month. Given that an erroneous decision may expose those most vulnerable to renewed danger or persecution, system reforms must define the boundaries of ADM within the asylum determination process.

What is Automated Decision-Making?

ADM can be broadly defined as any automated system utilised in organisational decision-making. It does not necessarily involve artificial intelligence; ultimately, the system will “materially [affect] either the substance of a regulated decision or the process by which a regulated decision is made”. Although the scope of ADM in migration decisions is unclear, the Department of Home Affairs notes it uses AI tools, including ADM, to support operational functions, manage risks, and perform administrative tasks.

The Nature of Decisions

Despite these uses, ADM may present genuine risks to vulnerable groups. Regulation (EU) 2024/1689 classifies AI systems used in the examination of applications for asylum, visa and residence permits as “high-risk”. It outlines that applicants are “often in particularly vulnerable position[s]” and are “dependent” on decisions by public authorities to exercise their fundamental rights.

University of Wollongong’s Niamh Kinchin highlights that refugee cases are often “shaped by emotion and uncertainty”, something ADM systems notoriously struggle to grasp. ADM’s inability to reason emotionally is particularly concerning, as many asylum seekers may struggle to recount traumatic events within a foreign legal system, often in a foreign language. The lack of such reasoning could also lead to assessments in which factors, including risks and benefits, are considered disproportionately. Errors in ADM results can make the practice dangerous for applicant triage, as mistakes at preliminary stages could cause delays, adding to processing times and backlog.

The 1951 Convention Relating to the Status of Refugees outlines a number of fundamental rights possessed by those seeking asylum. Most notably, these include the right to non-discrimination, a right which ADM’s well-documented problems with bias could threaten. Experts warn that systems trained on imbalanced data may systematically prefer certain decisions, such as producing unfavourable visa outcomes for individuals of certain nationalities, without sufficient consideration of their personal circumstances. As a result, ADM practices used by public authorities may unintentionally conflict with international regulatory frameworks on the rights of refugees and asylum seekers.

With several European countries discussing systems for offshore deportation and detention, seemingly influenced by the Australian approach, the stakes for accurate decision-making and policy have never been higher.

Equitable Outcomes

Although ADM systems may bring tangible benefits to processing times, a lack of human oversight risks producing inequitable outcomes. The implementation of ADM in foreign countries may provide Australia with a point of reference. For example, the UK Home Office’s use of ADM for triaging visa applications faced criticism amid concerns about potential bias based on applicants’ nationalities. After rights groups challenged it as discriminatory, the tool was discontinued in 2020, highlighting how difficult it is to make such a process completely fair. For Australia to utilise ADM in applicant triage, it must be unequivocal that the system is free from actual or perceived bias.

Another policy framework to consider comes from Norway’s Directorate of Immigration, which uses ADM in limited cases and only with consent from applicants. Given the complexities of ADM, Norway reserves these systems for residency applications for family migration of skilled and posted workers, and citizenship applications, and fully automates only those that receive a positive response. If Australia sought to mirror such a framework for protection findings, effectively mitigating the risks of ADM would likely mean ensuring only positive responses could be subject to minimal human oversight.

Although the Directorate has noted challenges with low numbers of consenting applicants, the use of ADM has shortened processing times. However, the vulnerability of some asylum seekers, however, raises questions about whether such a system, if used in humanitarian visa assessments, could be rooted in genuine consent.

In asylum matters, ADM is only used in the Norwegian framework for the distribution of welfare benefits and allocating reception centres best equipped for individual needs. Reception centre staff, as well as individuals, can seek to have these decisions reviewed and corrected when needed. Importantly, this means that the onus of correcting errors does not fall solely on asylum seekers, who may otherwise struggle to challenge decisions. By confining ADM to lower-risk tasks, such as service matching rather than primary visa decisions, and mandating options for review, Australia could reduce the risks associated with ADM.

The Case for Cautious Automation

Emerging guidelines should define when it is and is not suitable to utilise ADM. These should consider existing legal frameworks and include examples of what good uses would look like. Systems adopted for public administration should be designed to function with varying degrees of autonomy, with particular safeguards for vulnerable applicants. At minimum, these safeguards should mandate a right to review automated decisions, ideally extending a responsibility to flag incorrect decisions. Given how public authorities engage with often vulnerable people, carefully implemented ADM may not only increase the efficiency of administrative processes, but also deliver fairer outcomes for applicants.

Amrita Chahal is a recent graduate of the University of New South Wales, where she completed a Bachelor of Laws and Bachelor of Media (Communications and Journalism) with Distinction. She holds a Certificate in Transnational Law from the University of Geneva, strengthening her understanding of Australia’s role in international affairs.

Amrita has professional experience in law, public policy and international engagement. She serves in the U.S. Consulate General Youth Advisory Council. Additionally, Amrita has worked in refugee law, youth justice and human rights, shaping her interest in considering regulatory frameworks and international relations through a humanitarian lens. She is particularly interested in exploring how Australia can support the responsible global governance of emerging technologies.

Disclaimer: The views and opinions expressed in this article are those of the author, and do not necessarily reflect the views and opinions of Young Australians in International Affairs. AI tools were used by this author for grammar checks, but all content is original, and no plagiarism has been used in the preparation of this article.