The COE Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law. Is the Council of Europe losing its compass ?

by Emilio DE CAPITANI

When the Committee of Ministers of the Council of Europe decided at the end of 2021 to establish the Committee on Artificial Intelligence (CAI) with the mandate to elaborate a legally binding instrument of a transversal character in the field of artificial intelligence (AI), such initiative created a lot of hopes and expectations. For the first time, an international convention ‘based on the Council of Europe’s standards on human rights, democracy and the rule of law and other relevant international standards’ would regulate activities developed in the area of AI.  

The mandate of the CAI was supposed to further build upon the work of the Ad Hoc Committee on Artificial Intelligence (CAHAI), which adopted its last report in December 2021, presenting  ‘possible elements of a legal framework on artificial intelligence, based on the Council of Europe’s standards on human rights, democracy and the rule of law’. In this document, the CAHAI underlined the need for the future instrument to ‘focus on preventing and/or mitigating risks emanating from applications of AI systems with the potential to interfere with the enjoyment of human rights, the functioning of democracy and the observance of the rule of law, all the while promoting socially beneficial AI applications’. In particular, the CAHAI considered that the instrument should be applicable to the development, design and application of artificial intelligence (AI) systems, ‘irrespective of whether these activities are undertaken by public or private actors’, and that it should be underpinned by a risk-based approach. The risk classification should include ‘a number of categories (e.g., “low risk”, “high risk”, “unacceptable risk”), based on a risk assessment in relation to the enjoyment of human rights, the functioning of democracy and the observance of the rule of law’. According to the CAHAI, the instrument should also include ‘a provision aimed at ensuring the necessary level of human oversight over AI systems and their effects, throughout their lifecycles’.

So, a lot of hopes and expectations: some experts expressed the wish to see this new instrument as a way to complement, at least in the European Union, the future AI Act, seen as a regulation for the digital single market, setting aside the rights of the persons affected by the use of AI  systems[1]. In its opinion of 20/2022 on the Recommendation for a Council Decision authorising the opening of negotiations on behalf of the European Union for this Council of Europe convention, the EDPS considered that it represented ‘an important opportunity to complement the proposed AI Act by strengthening the protection of fundamental rights of all persons affected by AI systems’. The EDPS advocated that the convention should provide ‘clear and strong safeguards for the persons affected by the use of AI systems’.

Alas, those hopes and expectations were quickly dampened by the way the negotiations were organised, and, above all, by the content of the future instrument itself.

1- the organisation of the negotiations: the non-member States leading, the civil society out

The objective to open the future instrument to States which are not members of the Council of Europe was with no doubt an excellent initiative, considering the borderless character of AI, and the need to regulate this technology worldwide. Indeed, as noted by the CAHAI in its above mentioned report ‘The various legal issues raised by the application of AI systems are not specific to the member States of the Council of Europe, but are, due to the many global actors involved and the global effects they engender, transnational in nature’. The CAHAI therefore recommended that the instrument, ‘though obviously based on Council of Europe standards, be drafted in such a way that it facilitates accession by States outside of the region that share the aforementioned standards’. So, yes on a global reach, but provided that the standards of the Council of Europe are fully respected.

However, the conditions under which those non-member States have participated in the negotiations need be looked at a little more: not only have they been part of the drafting group sessions unlike the representatives of the civil society, but it seems that from the start they have played a decisive role in the conduct of negotiations. According to a report published in Euractiv in January 2023[2], the US delegation opposed the publication of the first draft of the Convention (the ‘zero draft’), refusing to disclose its negotiating positions publicly to non-country representatives.

At the same time, the organisation of the negotiations has set aside the civil society groups, who were only allowed to intervene in the plenary sessions of the meetings, while the text was discussed and modified in the drafting sessions. The next and-in principle- last plenary meeting from the 11th to the 14th of March should start with a drafting session and will end with the plenary session, which implies that the civil society representatives will have less than 24 hours to have a look at the revised version of the convention -if they can receive it on time- and make their last comments, assuming that their voices were really heard during the negotiations.

Yet, representatives of the civil society and human rights institutions have done their utmost to play an active part in the negotiations. In an email to the participating States, they recalled that the decision to exclude them from the drafting group went ‘against the examples of good practice from the Council of Europe, the prior practice of the drafting of Convention 108+, and the CoE’s own standards on civil participation in political decision-making”[3]. During the 3rd Plenary meeting of 11-13 January 2023, they insisted on being part of the drafting sessions, but the Chair refused, as indicated in the list of decisions:

‘(…) –Take note of and consider the concerns raised by some Observers regarding the decision taken by the Committee at the occasion of its 2nd Plenary meeting to establish a Drafting Group to prepare the draft [Framework] Convention, composed of potential Parties to the [Framework] Convention and reporting to the Plenary.

– Not to revise the aforesaid decision, while underlining the need to ensure an inclusive and transparent negotiation process involving all Members, Participants and Observers and endorsing the Chair’s proposal for working methods in this regard’.[4]

Despite this commitment, the need of an ‘inclusive and transparent negotiation process’ has not been ensured in the light of the civil society statement of the 4th of July 2023, where again the authors ‘deeply regret(ted) that the negotiating States have chosen to exclude both civil society observers and Council of Europe member participants from the formal and informal meetings of the drafting group of the Convention. This undermines the transparency and accountability of the Council of Europe and is contrary to the established Council of Europe practice and the Committee on AI (CAI) own Terms of Reference which instructs the CAI to “contribute[…] to strengthening the role and meaningful participation of civil society in its work”.’[5]

The influence of non-member States has not been limited to the organisation of meetings. As detailed below, the American and Canadian delegations delegations, among others, threw their full weight behind the choice of systematically watering down the substance of the Convention.

2- A convention with no specific rights and very limited obligations

How should the mandate of the CAI be understood? According to the terms of reference, the Committee is instructed to ‘establish an international negotiation process and conduct work to finalise an appropriate legal framework on the development, design, use and decommissioning of artificial intelligence, based on the Council of Europe’s standards on human rights, democracy and the rule of law and other relevant international standards, and conducive to innovation, which can be composed of a binding legal instrument of a transversal character, including notably general common principles (…)’[6].

The objective of including in the convention ‘general common principles’ has been interpreted by the Chair literally, who considered that ‘the AI Convention will offer an underlying baseline of principles in how to handle the technology, on top of which individual governments can then build their own legislation to meet their own specific needs’[7]. Indeed, the last publicly available version -dated 18 December 2023- of the draft Convention only refers to ‘principles’ and not to specific rights[8], even those already existing in the framework of the Council of Europe and beyond. In the context of AI, though, one could have hoped the recognition of certain rights, as the right to human oversight and the right to explanation for AI based decisions.

Such a choice has been criticized by the civil society‘s representatives. In a public statement of the 4th of July 2023, they recalled that ‘while including general common principles for AI regulation as indicated in the CAI Terms of Reference, the Convention should respect the rights established by other Conventions and not reformulate them as mere principles’[9].

Unfortunately, the Convention, at least in the version of the 18th of December 2023, does not even expressly include the right to privacy and the right to the protection of personal data. Yet, if data are, as the Chair himself referred to, ‘the oil of the XX1st century’[10], the need to protect our rights in this area is critical.

If one compares the successive versions of the Convention which are publicly accessible, from the zero draft[11], to the version of the 18th of December, one can only deplore the constant watering down of its content. What about ‘prohibited artificial intelligence practices’ referred to in Article 14 of the zero draft? What about the definitions, which included in the zero draft the notion of ‘artificial intelligence subject’, defined as ‘any natural or legal person whose human rights and fundamental freedoms, legal rights or interests are impacted by decisions made or substantially informed by the application use  of an artificial intelligence system’? What about a clear presentation of the risk-based approach, with a differentiation of measures to be applied in respect of artificial intelligence systems posing significant and unacceptable levels of risk (see articles 12 and 13 of the zero draft)?

Moreover, in the version of the 18th of December 2023, a number of obligations in principle imposed on Parties might become simple obligations of means, since the possible -or already accepted- wording would be that each party should ‘seek to ensure’ that adequate measures are in place. It is in particular the case in the article dedicated to the ‘integrity of democratic processes and respect for rule of law’, as well as in the article on ‘accountability and responsibility’ and even in the article on procedural safeguards, when persons are interacting with an artificial intelligence system without knowing it.

According to an article published in Euractiv on 31 Jan 2024 and updated on 15 Feb 2024, even the version of the 18th of December 2023 seems to have been watered down: ‘Entire provisions, such as protecting health and the environment, measures promoting trust in AI systems, and the requirement to provide human oversight for AI-driven decisions affecting people’s human rights, have been scrapped’[12].

3- The worse to come?

One crucial element of the Convention still needs to be discussed: its scope. Since the beginning of the negotiations, the USA and Canada, but also Japan and Israel, none of them members of the Council of Europe, have clearly indicated their wish to limit the scope of the instrument to activities within the lifecycle of artificial intelligence systems only undertaken by public authorities[13]. Moreover, national security and defence should also be out of the scope of the convention.  The version of the 18th of December includes several wordings regarding the exemption of national security, which reflect different levels of exemption.

The issue of the scope has lead the representatives of the civil society to draft an open letter[14], signed by an impressive number of organisations calling on the EU and the State Parties negotiating the text of the Convention to equally cover the public and private sectors and to unequivocally reject blanket exemptions regarding national security and defence.

Today no one knows what the result of the last round of negotiations will be: it seems that the EU is determined to maintain its position in favour of the inclusion of the private sector in the scope of the Convention, while the Americans and Canadians might use the signature of the Convention as blackmail to ensure the exclusion of the private sector.

4- Who gains?

From the Council of Europe perspective, which is an organisation founded on the values of human rights, democracy and the rule of law. the first question that comes to mind is what are the expected results of the ongoing negotiations. Can the obsession to see the Americans sign the Convention justify such a weakened text, even with the private sector in its scope? What would be the gain for the Council of Europe and its member States, to accept a Convention which looks like a simple Declaration, not very far in fact from the Organisation for Economic Co-operation and Development’s Principles on AI[15]?

At this stage, it seems that neither the Americans nor the Canadians are ready to sign the Convention with the inclusion of the private sector, even if an opt-out clause were inserted in the text. The gamble of the Chair and the Secretariat to keep these two observer States on board at the price of excessive compromises might be lost at the end of the day. One should not forget that these States do not have voting rights in the Committee of Ministers.

The second question that comes to mind is why the Chair and the Secretariat of the CAI and, above them, those who lead the Council of Europe have made such a choice. Does it have a link with internal decisions to be taken in the next future, as regards the post of the General Secretary of the organisation, as well as the post of the Director General of Human Rights and Rule of Law? Does the nationality of the Chair have a role to play in this game? In any case, the future Convention might look like an empty shell, which might have more adverse effects than it seems prima facie, by legitimizing practices around the world which would be considered incompatible with the European standards.

NOTES

---

[1] See in particular ‘The Council of Europe’s road towards an AI Convention: taking stock’ by Peggy Valcke and Victoria Hendrickx, 9 February 2023: ‘Whereas the AI Act focuses on the digital single market and does not create new rights for individuals, the Convention might fill these gaps by being the first legally binding treaty on AI that focuses on democracy, human rights and the rule of law’. https://www.law.kuleuven.be/citip/blog/the-council-of-europes-road-towards-an-ai-convention-taking-stock/

[2] https://www.euractiv.com/section/digital/news/us-obtains-exclusion-of-ngos-from-drafting-ai-treaty/

[3] same article

[4] https://rm.coe.int/cai-2023-03-list-of-decisions/1680a9cc4f

[5] https://ecnl.org/sites/default/files/2023-07/CSO-COE-Statement_07042023_Website.pdf

[6] https://rm.coe.int/terms-of-reference-of-the-committee-on-artificial-intelligence-cai-/1680ade00f

[7] https://www.politico.eu/newsletter/digital-bridge/one-treaty-to-rule-ai-global-politico-transatlantic-data-deal/

[8] with the exception of ‘rights of persons with disabilities and of children’ in Article 18

[9] https://ecnl.org/sites/default/files/2023-07/CSO-COE-Statement_07042023_Website.pdf

[10] https://www.linkedin.com/pulse/data-oil-21st-century-ai-systems-engines-digital-thomas-schneider/

[11] https://www.statewatch.org/news/2023/january/council-of-europe-convention-on-artificial-intelligence-zero-draft-and-member-state-submissions/

[12] https://www.euractiv.com/section/artificial-intelligence/news/tug-of-war-continues-on-international-ai-treaty-as-text-gets-softened-further/

[13] same article

[14] https://docs.google.com/document/d/19pwQg0r7g5Dm6_OlRvTAgBPGXaufZrNW/edit

[15] https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449