Author: Snehasree Parida, School of Law, Centurion University
Abstract
The recent and accelerating proliferation of AI to govern the public in criminal justice, employment, border control and other social service has reshaped the old paradigm of international human rights. This article raises this fundamental structural issue: Are the current international human rights regimes capable of addressing the systemic, black-boxed and cross-border impacts of algorithmic decision-making? Using instructive legal instruments such as the UDHR, ICCPR, and ICESCR, this article identifies the key normative and accession contradictions between them; then discusses the implications of predictive policing, biometric data collection and automated administrative governance for the right to privacy, right to non-discrimination, right to due process and right to fair trial; before analysing the state of emerging domestic and international case law and legislation, including the EU AI Act. It concludes that institutional and participatory design flaws emerge from their moribund moral and historical discourses, while their rules of enforcement need procedural and statutory modifications and shall have structural enhancements; and suggests a design to govern an international human rights regime to control the intervening power of AI institutions of state and non-state.
To the Point
The uncanny fact that the application of AI is happening at the same time as its exclusionary effects provides what amounts to a profound and broad structural crisis affecting the entire regime. The thesis answer is unexceptional. International human rights law, despite its intellectual strength, is functionally and constitutionally unable to tolerate monothe algorithmic trespass. The post-World War II international human rights system proceeds with an assumption that causality is anthropocentric and perhaps even ontological. In contrast, for AI to disorder, it generates autonomous, non-human, and non-transparent algorithmic entities that can break down linear chains of causality.
The most fragile in ever-fast technological advances seem to break down at the three main operational levels:
1) Administrative automation seamlessly moves its voice behind the due process: AI empowered administrative decisions without human intervention and fact-based explanation and calculation;
2) Properly intelligent prediction neural networks have systematically forced the global police and community onto the trajectory that the technology cannot be less and not even less developed than the pure mathematics at scale;
3) The corporation states union armed with the historical agencies of immense technologically efficient social administration of public and private sectors to the computers in data. In the course of regulatory aspiration, the EU regional all regulation, such as the European Union AI Act, suffers from miserably yet sadly hopeless neglect to the cause of dismal implementation record of international law at the supranational stage, especially for non-states and through the cross-border information flow.
Use of Legal Jargon
Clear language should be used in the discussion of AI and human rights. As AI is often opaque/black box, there are significant issues of explainability and transparency. Algorithmic bias or can be seen as discrimination or disparate treatment under Article 26 of the International Covenant on Civil and Political Rights (ICCPR), on Equality before the law and Article 14 of the European Convention on Human Rights (ECHR) on Equality and Non-discrimination. Automated decision-making, allowing a decision without human intervention, may infringe the right to a fair trial/Due process set out in Article 14 of the ICCPR. Novel concepts pertinent to data protection (e.g., CEEA Convention 108+) include proportionality, purpose limitation, and data minimisation. The non-justiciable personality of AI complicates the imposition of responsibility, relying on state responsibility in relation to the principle of due diligence (Velasquez Rodriguez v. Honduras). The prohibition on refoulement under art 3 ECHR shall be activated by the use of immigration AI algorithms that wrongfully deny asylum seekers. The margin of appreciation may allow a state discretion to choose between security and privacy, but it must be proportionate. Positive obligation also constrains states from allowing harm by private actors.
The Proof
This inability of the international human rights framework to effectively safeguard people against the partiality inherent to algorithms is, at least in some cases, supported by the doctrinal and empirical analysis. The maps of the International Bill of Human Rights (in treaties adopted through ratification by states) displayed numerous glaring omissions of practice.
1) The Gap Between Principle and Practice:
These most recent human rights norms were created in a world that did not yet have autonomous algorithmic systems. The UDHR (1948) and the International Covenant on Civil and Political Rights (1966) affirm the right to privacy (ICCPR, art.17), non-discrimination (art.26), and the right to a trial (art.14), but in each of those articles, the authors anticipated a human actor capable of being interrogated, reasoned with, or brought to account. The majority of AI is proprietary technology incapable of providing adequate justifications (at least at the current state of design). The human rights committee has not released a General Comment on AI, so there is a hodge-podge of opinions.
2) Discriminations by Design
Algorithms possess an inherent ability to perpetuate and amplify exclusion and discrimination. For example, AI- based predictive policing has been shown to affect minority groups disproportionately, which may violate the rights enshrined in Art 2 and 26 of the ICCPR. The high-level expert groups on AI convened by the EU has documented various instances of racial or sex discrimination by AI recruiting devices. Existing equality legislation like the Race Equality Directive ( 2000/43/EC) in the European Union, often does not have the evidentiary systems to apply the ill-effects test and identify disparate impact attributable to a secret algorithm. According to the Convention on the Elimination of All Forms of Racial Discrimination, states are obligated to prohibit racial discrimination in all its forms, but states may inadequately regulate private AI developers, thereby failing to fulfil their positive obligations.
3) Privacy and Surveillance:
The adoption of facial recognition by authorities raises serious privacy implications. R (on the application of Bridges) v. South Wales Police (2020) UK Court of Appeal held that the Police’s use of FRT was a violation of article 8 of the European Convention on Human Rights (ECHR) for the “lack of adequate safeguards in law”. However, such reasoning operates within domestic law. Internationally, the UN special Rapporteur on the right to privacy called for a moratorium on public-facing FRT; presently, no binding treaty is in place. The International Covenant on Civil and Political Rights states that any intrusion into the right must be “lawful, necessary and proportionate.” Most of the state’s rules providing for surveillance are not, especially considering the need for AI to both process and analyze otherwise ‘indiscriminate’ volumes of data.
4) Accountability and remedy
The right to an effective remedy is under article 2(3) of the ICCPR; however, the efficacy of such a right is in doubt when an AI system produces an incorrect result, such as refusing a loan application or misidentifying a suspect in the eyes of the law. Who is liable, the developer, the deployer or the state? The procedural element of rights has been eroded by the opacity of the algorithms used. In State v. Loomis (2016) the Wisconsin Supreme Court validated the use of risk assessment software COMPAS, despite the defence not being able to challenge its proprietary source code, leaving an interesting question mark hanging over liability. The US Supreme Court declined to hear the case, a good example of how the protection of trade secrets can outweigh the guarantee of due process and the ability to cross-examine a witness. On the international stage, although the European Court of Human Rights has not ruled directly on the use of AI, its jurisprudence on positive obligations and effective investigations, in cases such as M. C. v. Bulgaria (2003), shows there is an obligation on states to ensure that victims of AI-related abuses have access to a remedy.
5) Emerging Soft Law and Hard Law Efforts
The Ad Hoc Committee on Artificial Intelligence (CAHAI) of the Council of Europe has drafted a possible convention on AI, human rights, democracy, and the rule of law, to be finalized in 2024. While the EU’s AI Act establishes a risk-based regulation, it is, at its core, a trade regulation and not a Human Rights treaty. The UN Secretary-General’s Roadmap for Digital Cooperation highlights a global AI governance structure rooted in human rights, but does not propose one. None of these frameworks focuses on closing the enforcement gap. Without a dedicated international court or institution, victims must turn to national legal action or regional courts such as the ECtHR, which are overstretched and geographically limited.
Case Laws:
1) R (Bridges) v. Chief Constable of South Wales Police [2020] EWCA civ 1058 (UK)
Facts: Public areas within South Wales Police were monitored using Automatic Facial Recognition technology. The appellant argued the use of this technology was a violation of the right to privacy and data protection rights under the Human Rights Act 1998 (which incorporates Article 8 of the ECHR).
Held: The Court of Appeal found that the legal regulation governing the use of this technology was not of such certainty as to meet the quality of law standard set out in Article 8(2) of the ECHR. The lack of compulsory guidelines, the failure of data retention policies and the non-existence of an independent oversight body meant that the interference was disproportionate. Although not banning the technology outright, the court ordered that Parliament legislate with greater precision.
Significance: The case shows the existing principles of human rights (rule of law, necessity, and proportionality) are identifiable and applicable to AI, the evidence of extensive judicial analysis of this scenario is additional unique feature. In addition, it reveals another difference between reality and formal authorization.
2) State v. Loomis (2016) 881 N.W.2d 749 (Wisconsin, USA)
Facts: Loomis was convicted of six years for a past crime. An algorithm called COMPAS determined the risk assessment score that was calculated for Loomis. Loomis claimed that his use of a proprietary algorithm violated his due process rights because he could not challenge the methods used withthe algorithm.
Held: the Wisconsin Supreme Court affirmed, holding that the algorithm was merely one of several considerations and that revelation of the company‘s code would harm their business interest. They held to a recommendation of an addendum to presentence reports describing the disadvantages of risk assessment.
Significance: This case presents the problem of balancing between the secrecy of trade secrets and fair process. It might be at odds with Art 14 of ICCPR and art 6 of the ECHR, which guarantee the right to a fair trial, when the algorithm itself was dispositive. The ECHR might also find a breach of fair process according to its decision in Rowe and Davis v UK, (2000), allowing the defense an opportunity to confront its accusers.
3) Schrems II (Case C-311/18, CJEU, 2020)
Fact: Maximilian Schrems challenged the legality of the transfer of personal data from the EU to the US under the Privacy Shield, arguing that legislation governing US surveillance programs did not guarantee an adequate level of protection from intrusion by intelligence agencies.
Held: The Court of Justice of the European Union struck down the Privacy Shield Decision, stating US law failed to offer a level of protection that is essentially equivalent to that under the GDPR and the Charter of Fundamental Rights of the European Union (Articles 7, 8, 47), and emphasizing the importance of an effective remedy against surveillance.
Significance: While this case is not immediately relevant to AI, it sets a precedent for data protection regimes for AI systems. An AI model trained on data from within the EU, and used elsewhere in the world, could violate the ‘adequate level of protection’ requirement absent an adequate level of protection regime. A human rights-based approach leads back to the data itself.
4) Barbulescu v. Romania (2017) ECtHR (Grand Chamber) No. 61496/08
Facts: An employer who spies on an employee‘s private conversations through Yahoo Messenger. The Romanian courts found no breach of the right to privacy. The employeeargued this was a breach of Article 8 of the European Convention on Human Rights (ECHR).
Held: The Grand Chamber found that the national courts had failed to properly balance the consequences of the employer‘s interests against the employee‘s right to privacy. The Court pointed to the lack of specific, prior notice.
Importance: This case is significant in the context of AI-based workplace monitoring. The principles of proportionality and foreseeability of Barbulescu also apply to AI that records keystrokes, mood, or efficiency. States must make sure that their internal legal system provides ample protection, they will be violating their positive duty under Art 8.
Conclusion
The existing framework of international human rights law, while historically significant, now requires revaluation in the context of advancements in artificial intelligence (AI). Core principles such as non-discrimination, due process, privacy, and redress, as established by instruments like the ICCPR and ECHR, must be adapted to address the challenges posed by AI technologies. One significant issue is the “black-box” nature of AI, which can compromise procedural rights linked to human dignity. An example is the case of Loomis, illustrating the conflict between trade secrets and the need for due process, highlighting that legal requirements for algorithmic transparency in high-risk applications are inadequate.
Moreover, individual victims of AI-related rights violations often lack the means to challenge widespread systemic breaches. Only through collective redress can the scale of the problem be addressed, yet existing treaties do not provide mechanisms for such collective action. Additionally, violations may occur beyond national borders, complicating accountability.
To effectively respond to these challenges, a tripartite approach is necessary. This includes:
• Updating treaty body interpretations, specifically through a new general comment on AI from the UNHRC.
• Establishing a binding international treaty dedicated to AI, managed by the Council of Europe or the UN, which would include a robust complaints procedure.
• Implementing a national framework adhering to a ‘human rights by design’ paradigm, which encompasses transparency registers, impact assessments, and independent monitoring of high-risk AI systems.
Without these measures, human rights frameworks risk becoming obsolete, serving as an inadequate defense against the complexities introduced by digital technologies. It is crucial to prioritize the human rights of both individuals and vulnerable algorithmic subjects in the governance of AI systems, ensuring new legal frameworks are proactive rather than reactive to evolving realities.
FAQs
Q1. How can an individual seek a remedy for an AI-caused violation?
Ans. Usually through domestic courts or regional human rights bodies. The victim must exhaust domestic remedies first. The UNHRC can hear individual communications if the state has ratified the optional protocol to the ICCPR. However, proving causation and identifying the responsible party in an AI chain can be extremely difficult.
