The Convention on AI and Human Rights that no one talks about

A proliferation of international and regional, legal, ethical, and policy frameworks on AI saw the light last year. These frameworks are evolving faster than an AI-system can process and you might have missed some of them. Perhaps most notably, you might have missed an important framework Convention. That is because this Convention, despite its far-reaching implications, has not received the attention it deserves: a legally binding Human Rights Convention addressing the impact of AI on Human Rights, Democracy and the Rule of Law is currently in the final stages of negotiation within the Council of Europe. The latest draft text was unveiled shortly before Christmas 2023. When the text is finalized – which is scheduled for March 2024 -, this will mark a critical milestone in the AI and human rights governance framework.?

This article will provide you with essential background information on the Convention through the eyes of the person negotiating the Convention on behalf of the European network of National Human Rights Institutions. I will offer insights in the workings of the Convention and summarize the latest common position of Europe’s National Human Rights Institutions on the draft Convention. After shedding light on the Council of Europe and reminding what National Human Rights Institutions are, we will delve in the following 6 key recommendations. Where relevant, I will draw comparisons to the EU AI Act.

KEY RECOMMENDATIONS

1. Ensure a broad scope of application - Ensure the Convention applies to the public and private sector - Limitations related to the use of AI by law enforcement, national security?and defense should be in accordance with human rights law and standards - Remove unnecessary qualifiers, vague terminology and repeated references to domestic law
2. Ensure strong independent oversight at national and Council of Europe level
3. Address prohibited practices adequately
4. Address collective and societal-level harms?and ensure meaningful multi-stakeholder participation
5. Strengthen transparency requirements
6. Ensure consistent reference to internationally established 'rights' and ensure the right to an effective remedy

This article reflects my personal opinion and is no official position of ENNHRI.

Introduction

A Council of Europe Framework Convention on Artificial Intelligence, Human Rights, Democracy, and the Rule of Law, c’est à dire?

The Artificial Intelligence (AI) Convention is not the EU AI Act. While the EU AI Act focuses on establishing a comprehensive framework for the regulation of AI within the internal market of the European Union, the AI Convention takes a broader approach. The Convention aims to create an international binding treaty on AI and Human Rights, Rule of Law and Democracy.?

This initiative stands on a firm foundation of human rights protection. The Council of Europe (CoE), established after the Second World War in 1949, serves as a beacon for the protection of Human Rights, Democracy, and the Rule of Law across its 46 member states. With a mandate to foster collaboration and shared values, the CoE has played a pivotal role in shaping legal frameworks that uphold fundamental rights such as the European Convention on Human Rights, the 108+ Convention (nicknamed the global GDPR), the Budapest Convention (on Cybercrime), and the Oviedo Convention (on Biomedicine).

Amidst its historical commitment to human rights, the CoE is now spearheading efforts to create an international binding treaty on AI. The parties negotiating the Convention include a diverse array of global participants, demonstrating the CoE’s continued efforts and ambition to transcend regional boundaries. Alongside European nations, the participants include the European Commission, United States, Canada, Mexico, Japan, Israel, Ecuador, Peru, Uruguay and Argentina. You can find the list of participants via this link under plenary meetings.

Ensuring Human Rights in a Human Rights Convention: a challenging task anno 2024

In the rapidly evolving landscape of technology and human rights, national human rights institutions (NHRIs) struggle to carry out their mandate to protect human rights. NHRIs are established by constitution or law in accordance with the United Nations Paris Principles and the Council of Europe Committee of Ministers Recommendation 2021/1. These public and independent institutions have been created by States in all European countries and beyond as part of the public institutional infrastructure for human rights oversight. For the readers who are not familiar with NHRIs - in the words of the CoE - NHRIs are the bridge between government and civil society.

It is well-documented that enforcing human rights in the rapidly evolving technological era poses significant challenges, primarily attributable to the complex nature of technology, it's opacity and the diverse array of actors involved. This multifaceted environment introduces complexities that demand innovative and collaborative approaches to effectively address and uphold human rights. At this point, the European Network of National Human Rights Institutions (ENNHRI) and Network of Equality Bodies (Equinet) emerge as key advocates for human rights in the context of emerging technologies and AI.?

With over 40 NHRIs in Europe under its umbrella, ENNHRI is actively contributing to the Convention through its Observer status at the Council of Europe Committee on Artificial Intelligence (CAI). ENNHRI’s position on the draft framework Convention follows a previous position on the zero draft Convention, detailed written proposals and active participation as Observer to CAI sessions.

At this point, it is important to underline that different options for different key elements of the Convention are currently being proposed by the Secretariat. The following 6 points focus on the most important outstanding discussions from our perspective.

Key recommendations

1.??Ensure a broad scope of application

• Ensure the Convention applies to the public and private sector

The private sector is currently driving the development and deployment of AI. It is to that background rather strange that, the applicability of the Convention to the private sector is under discussion.

ENNHRI, together with other Observers and delegations, pushes for inclusion of?private actors in the scope of the Convention. The proposed exclusion of the private sector of the Scope of the Convention in article 3 (Options B and C) stands in sharp contrast with the adopted recommendations by CAHAI, CAI’s predecessor, which clearly established the scope of the Convention:

‘a transversal, binding legal instrument applicable to AI systems applied by all types of organisations, including public and private actors alike’.

Conventions impose general obligations to State Parties to respect certain rights and obligations and to ensure these generally apply to all individuals in their territory and subject to their jurisdiction, regardless of who the perpetrator is. The potential exclusion of the private sector from the scope of the Convention raises concerns as it will lead to a dual-track protection system. Moreover, it could establish a dangerous precedent for future human rights instruments. Imagine an AI system that causes discrimination. If this system is employed or developed by the government, the government must ensure the ? system is in accordance with the?principles and rights set out in the Convention, including providing for remedies (article 14). If the private sector would be the developer or deployer and the system is used within by a private entity or in a private space (e.g., a private school or hospital, a shopping mall, an apartment building, etc.), the provisions would not apply and such recourse may not be available.

Consider as an example scenarios involving major tech companies (including privately-owned social media platforms) and their potential influence on election processes: could citizens not rightfully expect that such situations fall within the purview of a binding Convention on AI, Human Rights, Rule of Law, and Democracy? Excluding the private sector from the Convention's scope prompts questions about the comprehensive protection of individuals in the face of advancing AI technologies.

?

• Limitations related to the use of AI by law enforcement, national security?and defense should be in accordance with human rights law and standards

ENNHRI is concerned by proposed exemptions related to national security in the scope of? draft framework Convention (Article 3).

On the specific issue of the use of AI by law enforcement and the potential exemptions or exceptions for national security and defense, drawing a comparison with the AI Act helps for a comprehensive understanding. Note that the legal foundation for the AI Act is Article 114 of the Treaty on the Functioning of the European Union (TFEU). The AI Act consequently aims to safeguard the internal market. However, as has been pointed out above, the mandate of the Council of Europe Committee on Artificial Intelligence (CAI) is more extensive, more precisely and expressly safeguarding human rights, rule of law and democracy.

The final text of the AI Act has not been released, and crucial details remain pending. However, civil society organisations including AlgorithmWatch, Center for Democracy & Technology (CDT) and the European Center for Not-for-profit Law (ECNL) have expressed concerns that the AI Act may fall short in adequately safeguarding fundamental rights because major loopholes exist (e.g. the 'subjective' filter in article 6 AI Act for risk classification), a blanket exemption for AI systems for national defense and national security and various exceptions for high-risk systems used in the contexts of law enforcement, and migration, where authorities can often avoid the reach of the Act’s core provisions.

Any limitation related to national security should be in accordance with the European Convention on Human Rights and international human rights obligations such as the International Covenant on Civil and Political Rights (ICCPR) and International Covenant on Economic, Social and Cultural Rights (ICESCR). A blanket exemption for national security is not in accordance with either the CoE standards and the jurisprudence of the European Court of Human Rights, nor with the ICCPR/ICESCR standards.

If it is necessary to clarify the limitations to the mandate of the Council of Europe in relation to national defense in the Convention, reference can be made to article 1d of the CoE Statute. The interpretation of how article 1d of the Statute of the CoE restricts the scope of a particular Convention is left to the interpretation of the European Court of Human Rights and of the other non-CoE-member States' courts where applicable.?

As discussions unfold around the AI Act, it is crucial to ensure that the CoE does not replicate the exemptions and loopholes present in the AI Act. Legislation should align with relevant human rights (or fundamental rights in EU terminology) standards. The AI Act falls short in this regard, possibly due to its legal foundation. There is a risk of a similar exemptions in the Convention, which is unacceptable given its higher, more stringent standards. The primary purpose of the Convention is to safeguard Human Rights. Consequently, there is a concern that, in neglecting these standards, the child risks being thrown out with the bathwater. Doing so will establish a dangerous precedent, diverging from the established CoE standards on the protection of Human Rights.

• Remove unnecessary qualifiers, vague terminology and repeated references to domestic law?

ENNHRI proposes to review and remove unnecessary and subjective qualifiers that do not provide any legal certainty or reasonable predictability.?

The draft also contains imprecise language that unnecessarily dilutes commitments, such as four times 'seek to ensure (/protect)' in crucial provisions (articles 3, 5 10, 12, 15) and two times ‘where practicable, to facilitate/promote’ in article 26. The ambiguity in such terminology raises concerns about the enforceability and effectiveness of the obligations outlined in the Convention. Clearer and more definitive language is necessary? to strengthen the language of the Convention, ensuring legal certainty and a level playing field.

ENNHRI further cautions against the repeated reference to ‘in accordance with domestic law’ These references may render crucial elements of the Convention dependent on the domestic law of the Parties. It is commendable that the Convention allows for flexibility to accommodate national specificities. However, it is important that such references to domestic law complement the aim to ensure common international legal standards on AI, human rights, democracy and rule of law rather than undermine it.? It is advisable to limit the references to domestic law to ensure the Convention ensures compliance with existing Council of Europe standards on human rights, the rule of law and democracy.?

?2 .?? Ensure strong Independent Oversight at National and Council of Europe Level?

ENNHRI emphasizes the importance of independent oversight and enforcement mechanisms at national and CoE level.

The following key elements should be included to ensure independent and effective oversight of the Convention:?

• The oversight mechanisms should have the necessary powers and as a minimum, these should explicitly? include investigative, complaint-handling, reporting and promotional powers.
• Explicit reference should be made to the need for formal and functional independence of the oversight authority, as well as adequate provision of resources to carry out its mandate, and a pluralistic composition reflecting multi-stakeholder participation. The Explanatory Memorandum could develop more specific safeguards for this independence, resources, and pluralism in line with similar provisions in the Council of Europe Convention 108+ as well as taking into account existing CoE independence standards for human rights supervisory authorities such as those from CM Recommendation 2021/1 and ECRI General Policy Recommendation N°2.?
• Meaningful cooperation with existing independent national human rights structures is key, including the possibility to trigger investigations, the duty to inform and consult when risks to the protection of human rights, the upholding of rule of law and the protection of democracy are identified.?
• When multiple bodies are appointed as supervisory authority under the Convention, the current article 26 (3) limits their cooperation and coordination to “taking measures – where practicable - to facilitate effective cooperation among them”. Given the complexity of AI and its oversight (covering action across different sectors, at different levels, and with possibly varied roles), the provision of a coordination mechanism would be warranted in case of multiple appointed oversight authorities. The United Nations Convention on the Rights of Persons with Disabilities (article 33(1)) already takes this approach and has been implemented in many countries across the world.

At CoE level, a Convention Committee (article 24), will be established. The composition should allow for sufficient expertise in the AI domain, comparable to the Committee set up under Convention 108+ which is composed of representatives of all Parties, ‘from the national supervisory authorities or from the government’. In order to enable adequate interaction between national supervision and the European monitoring mechanism, national supervisory authorities should at least be granted observer status within the Committee.?

Equinet and ENNHRI submitted a Joint statement on Chapter VII of the draft Convention to CAI, highlighting the lack of sufficient safeguards for the compliance of AI systems with human rights obligations in the current draft text, and proposing a series of amendments.?

?

3.???Address prohibited practices adequately?

The latest draft lays down a very general provision on possible bans of certain applications of AI systems, without any specific criteria or examples (Article 16 (3)). ENNHRI considers that the level of abstraction makes it very difficult, if not impossible, to ban certain AI applications posing unacceptable levels of risk, thus contradicting the broad consensus at CAHAI about the need for the future Convention to set certain ‘red lines’. ENNHRI submits that clear and unambiguous criteria, together with a list of specific prohibited uses of AI, are crucial in the context of the protection of individuals’ human rights and would be conducive to innovation whilst guaranteeing the necessary legal certainty and foreseeability for the developers, providers and users. Moreover, this approach would be in line with the EU AI Act, Council of Europe Convention on Human Rights and Biomedicine and its Additional Protocol on the Prohibition of Cloning Human Beings.?

Furthermore, it would be important to ensure the periodic review of these prohibitions in a manner that is sufficiently flexible. Currently, it is not specified how the incompatibility of certain uses of AI systems with the respect of human rights, the functioning of democracy and the rule of law would be assessed, and by which authorities. ENNHRI recommends specifying these powers shall be attributed to the oversight mechanism (article 26).?

?

4.???Address collective and societal-level harms,?and ensure meaningful multi-stakeholder participation

ENNHRI recommends to properly address the collective and societal-level harms that AI systems can cause, even when they cause only negligible harm to individuals. The Convention does not sufficiently take into account AI’s impact on interests with a more societal scope, which goes beyond impacts on individual interests, such as risks to the environment, language, culture, democracy, and rule of law. ENNHRI recommends including necessary procedural safeguards such as multi-stakeholder participation and representative rights, the right for individuals and associations to contest and seek redress for all AI systems that create individual, collective or societal harms (without the need to demonstrate individual harm) or infringe on other obligations set out in national and regional regulations on AI, and ensuring multi-stakeholder participation in enforcement on national and Council of Europe level.?

5.??Strengthen transparency requirements?

Given the importance of transparency and oversight in the context of artificial intelligence systems, the current formulation of the transparency requirements in the AI Convention should be strengthened. The current formulation is too abstract, and specific requirements should be included. ENNHRI recommends a public registry for AI systems and an obligation to report potential breaches of human rights.?

It is well-documented that enforcing human rights in the rapidly evolving technological era poses significant challenges, primarily attributable to the complex nature of technology and the diverse array of actors involved. This multifaceted environment introduces complexities that demand innovative and collaborative approaches to effectively address and uphold human rights. It is imperative that the relevant stakeholders have access to Human Rights Impact Assessments in order to comprehensively address the human rights implications. More specifically, National Human Rights Structures should have the power to access any documentation created or maintained under this Convention when access to that documentation is necessary for the fulfilment of the competences under their mandate, this recommendation is in line with article 64(3), 64 and 65 EU AI Act.?

6.????? Ensure consistent reference to internationally established 'rights' and ensure the right to an effective remedy?

Last but not least, the Human Rights Convention should set out Human Rights, not principles. And no, this is not a purely symbolic discussion. It is crucial that the Convention ensures consistent reference to internationally established rights, including the European Convention on Human Rights (ECHR), and does not reformulate rights as mere 'principles’. The current draft refers to ‘principles’ instead of rights. For example, the right to non-discrimination is internationally and universally recognised and should not be reformulated as a mere principle. This could lead the State Parties to interpret them as non-binding guidelines for action and raises difficulties in terms of enforcement. Reformulating rights as principles undermines the raison d’être of the Convention as indicated by its title.?

The latest draft includes a possibility for persons concerned to lodge a complaint, a welcome evolution. The right to an effective remedy encompasses the safeguarding of both individual and collective interests for those impacted by AI systems.?The comprehensive realization of the right to an effective remedy in the Convention should encompass the right to lodge a complaint, the right to human review, the right to be notified, and the right to meaningful explanation. Regrettably, the majority of these elements are still under discussion. The same applies to other important rights such as the protection of privacy and personal data (proposed article 10), the protection of whistleblowers (proposed article 19) and the preservation of the environment (addition proposed to article 11). The formulation and adoption of?draft articles 14, 15 and 19 is crucial for ensuring effective remedies in cases of AI-related impacts. This is in line with the recommendations put forth by CAIs predecessor CAHAI.

Is the Convention worthy of its title?

While the aspirations are lofty, ENNHRI's?latest common position sheds light on nuanced concerns and challenges that may cast doubt on whether the Convention is truly worthy of its ambitious title.

ENNHRI will continue to engage with CAI, submitting detailed text proposals and participating in upcoming drafting group meetings as an observer, in cooperation with CSO observers. Do not hesitate to reach out to join the discussion.