Law360 Canada (September 17, 2026, 11:26 AM EDT) --
 |
| Aavisa Butt |
For justice to exist alongside AI governance, we must adopt a sociotechnical approach instead of a purely technical approach to safeguard rights. This requires maintaining the justice system, including courts and other legal institutions, at the core of the sociotechnical system, revealing how language shapes law and governance, and how power, authority and control over AI remain subject to human responsibility.
Its value lies in addressing the contestation and indeterminate application of ethics in AI governance, integrating the justice system to establish procedural fairness, and recognizing the interpretativeness of language in shaping legal meaning and human reasoning. This means minimizing opaque linguistic understanding, lack of consequences and legal penalties, and delayed access to justice, ensuring that citizens can translate AI-related harm into legal claims capable of being examined and remedied.
We will rely on the postulate that society, as a whole, does not possess the sophisticated technical understanding of how AI operates at a code level, its limitations and black box problems. When the general population does not speak the “language” to understand how and why AI produces particular outputs, a linguistic gap emerges between the technical operation of AI and the people affected by its outputs. Where understanding is concentrated among those who create, deploy or control AI systems, they retain greater control over the narrative surrounding how the technology operates, what its outputs mean and how its risks are understood, placing those who cannot speak or understand this technical language at a disadvantage in challenging that narrative or exercising meaningful control.
Concerning access to justice, this linguistic gap becomes particularly significant when AI makes or assists decisions affecting rights and interests because a decision is insufficient if its reasoning cannot be understood or challenged. The courts and justice system therefore become essential actors within the sociotechnical system, providing the institutional and legal mechanisms through which AI-related decisions and harms can be scrutinized, challenged and remedied.
Additional actors outside the technical space, such as legal practitioners and policymakers, must therefore participate in interpreting and translating these technical processes into reasoning that can be understood, scrutinized and challenged.
To begin, the emphasis on the justice system establishes the need for procedural fairness. When it comes to AI governance and the technical approach, one of the dangers is lack of transparency stemming from lack of technical understanding. However, when we involve the justice system through a sociotechnical approach, we demand a fair and transparent understanding of what our rights are, what those rights entail, what the penalties are, what the legal tests are, and what policies and audits exist to ensure compliance with the justice system’s statutes and acts. Because the justice system relies on logical reasoning and context to proceed, argue and sentence, corporations cannot simply “hide” behind purely technical language or performance metrics; AI-related decisions remain subject to procedural and fairness requirements.
This is also where procedural fairness directly intersects with access to justice. If an individual cannot understand the basis upon which an AI-assisted decision has been made, their ability to challenge that decision is correspondingly weakened. Access to justice therefore requires more than the formal existence of legal procedures; individuals must be able to understand the decision affecting them and meaningfully participate in the process of challenging it. Transparency consequently becomes an important condition for ensuring that AI does not create an additional barrier between individuals and the justice system.
For example, suppose a mortgage application was denied by a bank following an AI-assisted assessment, without an open and transparent rationale being provided to completely explain why. That client then demands the rationale, but technical opacity prevents them from understanding the basis of the decision. Or the inability to obtain such an explanation can therefore become an access-to-justice issue. The bank’s rationale could itself contain social inequalities and feedback loops that perpetuate factors such as discrimination, meaning that the issue is not only whether the decision can be explained, but whether the reasoning underlying it is itself fair.
The linguistic gap represents not merely a deficit in technical understanding, but a distribution of power. Where technical actors retain greater control over how an AI system is explained, interpreted and justified, affected individuals may lack an equivalent ability to challenge the narrative surrounding the decision. Bringing the justice system into the sociotechnical system disrupts this imbalance by providing an independent institutional framework through which those explanations can be scrutinized against legal standards of fairness, rights and accountability.
A purely technical definition of fairness cannot be allowed to determine the legal outcome because the very actors responsible for maintaining the technical system may also be the actors whose practices need scrutiny to maintain integrity. For example, a technical approach measures fairness based on a computer scientist's performance metrics. Therefore, if any discrimination or infringement of rights arises from those performance metrics, though they may be more difficult to audit and thus enforce penalties upon, it does not mean that if the justice system were to hear a plaintiff's case, who has been discriminated against by those same metrics, to go unheard.
A sociotechnical approach, according to Oxford Academic’s
Policy and Society submission, “When code isn’t law: Rethinking regulation for artificial intelligence,” makes it easier and more transparent by utilizing a multidisciplinary perspective to include issues such as bias, discrimination and other structural inequalities. It allows the legal system to examine the social consequences of those technical processes and provide a forum in which the affected individual can be heard. Therefore, access to justice operates as a safeguard against technical complexity becoming an additional barrier to legal recourse.
This then gives rise to another important point: the definition of fairness is contested across disciplines. In the world of responsible AI governance, there is no single agreed-upon definition of “fair” AI. And those definitions are often in conflict with each other. An AI system might be considered fair by an app developer, but unfair by a judge. An app developer may have metrics based on a fair technical performance, whereas a judge would challenge those same metrics based on judicial transparency. A sociologist may further examine the social contexts in which AI systems are embedded, arguing that an AI system may appear ostensibly fair based on fairness metrics, yet reproduce existing societal inequalities, revealing it to be unfair.
However, when we involve the judicial system, we involve a key player in the system that has the expertise, resources, and impartiality to define, on the merits of a case and the interests of civil society, what fairness is and how it was violated. The judicial system can involve expert witnesses to explain AI systems and impose penalties. We can also cut through the ethical interpretativeness of what “fairness” means by allowing legal practitioners, policymakers and other qualified professionals to engage. By bringing judicial and multidisciplinary scrutiny into the sociotechnical system, control over the meaning and consequences of AI is no longer confined to those who design or deploy the technology.
Yet identifying and interpreting fairness is not enough if those determinations cannot produce legally enforceable consequences. Hard law and soft law are key to understanding the debate surrounding their relevance when faced with the rapid innovation of AI. In “
Bridging soft and hard law in AI governance,” Marco Pasqua states, “In AI governance … soft law’s utility is particularly evident in fast-evolving technological domains where flexibility is key: unlike hard law, which requires lengthy legislative processes.”
Derived from
Nature’s “The global landscape of AI ethics guidelines” (pp. 389-399), while it is true that most guidelines are voluntary and not legally binding, the problem arises when companies can claim to follow ethics without legal consequences. There is the factor of the dominant corporate influence as many guidelines are created by tech companies and industry groups. Here, the core issue then becomes a conflict of interest: companies may shape ethics to protect themselves, leading to regulatory capture.
This is where the approach also offers the possibility of combining hard law and soft law to create a dynamic regulatory framework. Hard law empowers the judicial process itself to exert punishment by widening the scope of the definition of “fairness.” It transfers authority from voluntary technical or corporate standards towards legally recognized institutions capable of determining whether AI practices comply with rights and imposing consequences where they do not. The distinction between soft and hard law is therefore relevant to whether access to justice can produce meaningful consequences rather than merely recognizing that harm has occurred.
Derived from the
Oxford Handbook of Ethics of AI (Chapter 11), in AI governance, there exists the problem of many hands. This is the “whole-system approach.” AI systems involve developers, companies, data scientists, users and end users. Naturally, when harm happens, responsibility is unclear. The main concern of responsible AI governance is thus the identification of the relative responsibility of all actors involved in the design, development, deployment and use of AI systems.
In addition, the sociotechnical approach empowers accountability as public authorities may legally have to disclose information, explain decisions and ensure accountability. The inclusion of these multiple stakeholders is also crucial for anticipating and addressing social implications, preventing the exacerbation of existing inequalities and resolving tensions. By holding multiple stakeholders accountable for the development and use of AI, a participatory approach can promote greater fairness in the justice system and allow for greater scrutiny and oversight of AI technologies used in court adjudication.
In “
Open justice and technology: Courts, tribunals and artificial intelligence,” Monika Zalnieriute states, “Automated systems generally do not (and possibly cannot) provide reasons for the decision they deliver,” while reasons are essential for ensuring that parties and the public understand the logic behind judicial decision-making. It emphasizes that language is a constitutive element of legal judgments, raising the question of whether code can provide reasons that affected individuals can actually understand, challenge and control.
This makes interpretation particularly important to access to justice. If citizens cannot understand the language through which an AI system’s operation or decision is communicated, formal legal rights may exist without being practically accessible to those affected. Meaningful access therefore depends upon the ability to translate technical processes into forms of reasoning that individuals and legal institutions can understand and scrutinize. An individual cannot meaningfully challenge a decision if they cannot understand the reasoning upon which it was based, particularly where that decision affects their rights or interests.
Those that command AI are, inevitably, human, and where human actions are concerned, the law gazes upon them with an eye of scrutiny. AI may exercise technical power, but the authority to design, deploy, govern and rely upon these systems ultimately remains with human actors and institutions. Procedural fairness and legal frameworks of the judicial process remain essential for determining who holds power, who exercises control and who bears responsibility when AI contributes to harm. Justice, therefore, represents the point at which transparency, procedural fairness, power, authority, accountability and legal enforceability converge, ensuring that the use of AI does not place those who exercise power through it beyond the reach of law.
Aavisa Butt is pursuing a master’s in artificial intelligence with a specialization in criminal justice systems. She holds an honours bachelor’s degree in English literature and an honours paralegal diploma.
The opinions expressed are those of the author(s) and do not necessarily reflect the views of the author’s firm, its clients, Law360 Canada, LexisNexis Canada or any of its or their respective affiliates. This article is for general information purposes and is not intended to be and should not be taken as legal advice.
Interested in writing for us? To learn more about how you can add your voice to Law360 Canada, contact Analysis Editor Peter Carter at peter.carter@lexisnexis.ca or call 647-776-6740.