United Nations, artificial intelligences and regulations: analysis of the General Assembly AI Resolutions and the Final Report of the Advisory Body on AI⋆

Abstract

Artificial intelligences are generating profound changes in our societies. They have been part of our daily lives for years: they are omnipresent, powerful, opaque and, many times, invisible to the human eye. Hence, its regulation is one of the most relevant topics within the technological agendas at the national, regional and international level. In 2024, the AI reached a historical peak of interest and discussion by the United Nations. Firstly, were adopted two resolutions by the General Assembly: Resolution 78/265 “Seizing the opportunities of safe, secure and trustworthy artificial intelligence systems for sustainable development” and Resolution 78/311 “Enhancing international cooperation on capacity-building of artificial intelligence”. Secondly, and related to UN Secretary-General, was published the Final Report from the Advisory Body on Artificial Intelligence “Governing AI for humanity”. Among other questions that this short paper seeks to answer: what are the most relevant points of these new regulation on AI? What are their proposals and recommendations? The objective of this paper is to describe and analyze the proposals of these international instruments. This contribution is part of a broader research analyzing AI policies and regulations worldwide.

Keywords

• eol>artificial intelligence
• United Nations
• sustainable development
• human rights1

1. Introduction: United Nations and its “tactical” position on AI

Artificial intelligences (henceforth AI) are generating profound changes in our societies [ 1 ] [ 2 ]. These have been part of our daily lives for years: these are different technologies that could be defined as omnipresent, powerful, opaque and, many times, invisible to the human eye [ 3 ]. AI are becoming access to information/data, value creation or knowledge management. For years, the United Nations (henceforth UN) has been warning about the acceleration of technological change, a kind of exponential growth, and how it could affect the world's population. However, despite the warnings, the position of the UN and its agencies about AI remains ambiguous. On the one hand, they place great hopes that digital technologies and, in particular, AI, can be the tactical tools that will allow us to achieve the 17 Sustainable Development Goals (henceforth SDGs) by 2030 [4]. On the other hand, they warn about the risks that these technological changes entail for human rights: among others, asymmetries in access/availability, digital divides, monopolies, violation of privacy and protection of personal data, data extractivism, algorithmic biases, discrimination, fake news, disinformation, autonomous weapons, etc. [5] [6]. Hence, AI regulations have become a recurring topic within UN and its agencies, reaching in 2024 a historical peak of interest and discussion. The objective of this short paper is describe and analyze the proposals of these new international instruments. In the next sections, two resolutions of the UN General Assembly and a Final Report of the Advisory Body on AI are analyzed.

2. General Assembly AI Resolutions: 78/265 and 78/311

On March 21, 2024, in the context of the 78th session, the United Nations General Assembly adopted a resolution on artificial intelligence, its regulation and the protection of human rights. The resolution was titled “Seizing the opportunities of safe, secure and trustworthy artificial intelligence systems for sustainable development” (A/78/L.49) [7]. The proposal was initiated by the United States (US) and, after several months of negotiation, it had the support of 123 countries. This is the first resolution of the General Assembly on AI issues: although it cites the classic precedents on the matter2, it was characterized by having no objections from China and Russia and by having been adopted by consensus3. For US Representative Linda Thomas-Greenfield, the main virtue of the Resolution is to open the dialogue on AI issues to the greater global community4. Despite the fact that the Resolution is easy to read and is translated into several languages, at times its wording becomes redundant, repetitive and circular. However the Resolution have, al least, five points that could be considered relevant to this research: 1) An initial definition of “safe, secure and trustworthy AI systems” (which could be classified as quite broad and not very operational)5; 2) An invitation to use the potential of AI to foster progress toward SDGs and the 2030 agenda (but without any empirical case to reference)6; 3) The option to reject and non-use (stop-using) AI systems that violate human rights (or put them at risk); 4) An invitation to Member States and multi-stakeholders to develop regulatory frameworks; and 5) A recommendation to take all necessary measures to bridge the artificial intelligence gap and other digital divides (between and within countries) [13]7.

Finally, the Resolution 78/311 was adopted by the General Assembly" on 1 July 2024, also without vote, and under the title "Enhancing international cooperation on capacity-building of artificial intelligence" [14]. The Resolution describes the position that UN is taking for the next years on AI capacity-building and international cooperation. There are here, at least, two relevant consideration related to IA politics and regulations. The first one is the attempt to use the international cooperation to solve AI divides and other digital divides between and within countries (including through North-South, South-South and triangular cooperation)8. The second 2 Its immediate antecedents include national and international initiatives: among others, Bletchley declaration, Global Partnership on AI (GPAI), Hiroshima Process signed by the G7, G20 principles for Trustworthy AI and OECD principles about AI. One of the objectives of the Resolution is to amplify the AI recommendations of UN agencies: among others, International Telecommunication Union (ITU), United Nations Educational, Scientific and Cultural Organization (UNESCO) [8], and Human Rights Council. 3 The adoption of a resolution by consensus means that the project presented was not submitted to a formal vote (with affirmative, negative votes or abstentions) but that no objections have been raised by the Member States [9]. 4 At a press conference, the US representative, together with representatives from the Bahamas, Japan, the Netherlands, Morocco, Singapore and the United Kingdom, considered that the Resolution is a great step to govern AI before these technologies govern us. Furthermore, he stressed that innovation and regulation are not mutually exclusive but complementary [10]. 5 The Resolution defines “safe, secure and trustworthy” AI systems as those that, belonging to the non-military field and covering their entire life cycle, they are characterized by: [a] human-centred, [b] reliable, [c] explainable, [d] ethical and inclusive, [e] fully respectful of human rights and international law, [f] privacy preserving, [g] sustainable development oriented and [h] be responsible [6]. Despite the abundance and diversity of the elements described, also present in several international documents, this is only an approximate, broad and ambiguous definition of these systems. A strict definition, so necessary in this field, could have led to explicit differences and delayed the adoption of the project by consensus. However, the Resolution warns of the urgency of reaching a global consensus on what these “safe, secure and trustworthy” AI systems are and how they would be developed. 6 The Resolution does not cite success stories about AI and the SDGs, nor other sources that allow us to identify how this virtuous articulation between digital technologies and the 2030 agenda will occur. It is important to mention that in 2023 the United Nations began a mid-term evaluation on the degree of progress of said objectives and the results were alarming [4]. The poor mid-term results and the world going through a polycrisis (pandemic, climate change and wars) forced United Nations expert groups to rethink the strategy and opt for intensive use of digital technologies and AI to accelerate results [11] [12]. 7 The Resolution recognizes that there are varying levels of technological development (between developed and developing countries) but does not provide data regarding these asymmetries (for example on AI academic/papers contributions worldwide [13]). 8 In particular, encourages Member States to increase capacity-building cooperation including all kind of activities: among others, policy exchanges, knowledge sharing activities and the transfer of technology on mutually agreed terms, technical assistance, lifelong learning, personnel training, skilling of workforce, international research cooperation, consideration is related to encourages Member States to consider the benefits and risks of opensource artificial intelligence (open-source software, open models and open data), digital public infrastructure and the use of digital public goods. These considerations seem to anticipate several of the proposals that would be published two months later in the Final Report of the Advisory Body on AI.

3. AI Advisory Body Final Report “Governing AI for humanity”

The multi-stakeholder High-level Advisory Body on Artificial Intelligence, initially proposed in 2020 as part of the United Nations Secretary-General’s Roadmap for Digital Cooperation (A/74/821), was formed in October 2023 to undertake analysis and advance recommendations for the international governance of artificial intelligence. Its first document was an interim report from December 2023 [ 1 ]. The second document, publicly presented in September 2024, was the Final Report titled "Governing the AI for humanity" [ 2 ]. In just over 100 pages, the Report is structured through an state-of-art of AI global governance and seven recommendations to improve it. In the initial section, without a doubt the most accurate and interesting, the Report describes several problems about AI global governance: among others, fragmented regulations, incipient efforts, voluntaristic norms, little effectiveness and lack of transparency. The Report express that the global level is the unique chance to reduce regulatory friction across borders. And, in particular, beyond the political and regulatory problems, the situation in developing countries is even more complex in terms of infrastructure and computing power for AI9. Beyond the “irrefutable” need to advance broad governance at a global level, the Final Report starkly shows the failures and gaps of this governance. A detailed reading of the report allows us to infer that UN and its agencies can do very little against the geopolitical positioning of some states and their corporate-commercial developments. The Final Report also presents, perhaps the most uncertain and pretentious section, seven recommendation (or solutions) that could strengthen IA governance for coming years: 1) An international and multidisciplinary scientific panel on AI (where members serve in their personal capacity on a voluntary basis); 2) An intergovernmental and multi-stakeholder policy dialogue on AI governance (on the margins of existing meetings at the UN); 3) AI standards exchange (with standard-development organizations and tech companies); 4) Capacity development network (UNaffiliated capacity development centers and AI training data); 5) Global fund for AI (managed by an independent governance structure that would receive financial/contributions from public and private sources); 6) Global AI data framework (developed by a relevant agency and formed by the work of other international organizations); 7) AI office within the UN Secretariat (drawing, wherever possible, on relevant existing United Nations entities). Beyond these characterizations and recommendations on AI governance, the Final Report expressly highlights something disturbing and paradoxical for an international instrument: the members of the Advisory Body have participated just in their personal capacity, not as representatives of Member States or their respective organizations10. Therefore, unlike the UN General Assembly Resolutions analyzed above, this Report does not have a high level of binding or commitment by the Member States of the United Nations.

4. Final thoughts: new needs and dangers of AI governance

The two General Assembly Resolutions adopted so far in 2024 indicate the interest that AI has for the UN and also the recognition that AI international governance issues and AI gaps need to be urgently addressed. These Resolutions had the virtue of having been adopted by consensus at a training courses, seminars and workshops [14]. 9 The report express that the computing power is one of the biggest barriers to entry in the AI field. None of the top 100 high-performance computing clusters in the world capable of training large AI models is hosted in a developing country [ 2 ]. 10 It is even stated that language included in this report does not imply institutional endorsement by the members' organizations [14]. time of growing global differences and strong geopolitical tensions over scientific-technological development. The adoption of these resolutions shows how the United Nations system, in accordance with its mandate, was able to reach some decisions (albeit precarious and somewhat ambiguous) about what could be defined as “safe, secure and trustworthy AI systems” and also issues on an "AI international cooperation". Unfortunately there are also several problems in relation to these international instruments. There are no common languages or standards among these resolutions and previous AI international instruments. So, in legal, ethical or technological terms, among others, it is very hard to define what is fairness, safety, explainability or transparency [15]. In the same way, many emerging standards are not grounded in a common understanding of meaning or are directly divorced from the values that they were intended to uphold. Perhaps the AI Advisory Body Final Report could have strategically helped resolve some of these issues, but its objectives were different. The expectations of researchers and academic communities about this report were high. However, despite its title, it seems more aimed at creating a superstructure on AI at UN than offers solutions toward the use AI for SDG and populations. Until now the Report is only an instrument that expresses a sum of individual positions with no binding capacity. At least the most relevant issue of the document is to recognize a serious AI governance problem and understand that it can only be resolved at the international level. Current geopolitical tensions could result in a world divided into disconnected and incompatible AI governance regimes. As has been recognized by the UN and its agencies, AI has existential implications. Unfortunately, the UN position on AI still remains dual: it seems to pendulum between the growing humanitarian needs of a world in polycrisis (pandemic, climate change and wars) and the evident violation of human rights behind the massive use of the AI. Actually, these technologies have an enormous potential but, far from being neutral, transparent or universal, are currently developed for profit, strong geopolitical interests and, so far, are not characterized by protecting human rights. One of the greatest dangers that AI entails is deepening global inequalities and concentrating them only on a handful of states and their corporations. The great challenge with AI, central to achieving SDGs (and, in particular, SDG No. 10), is whether its potential can be used to serve the common good and benefit all humanity. Until now, the UN instruments analyzed seem to be far from this challenge.

Acknowledgements

This work was supported by the ‘Consejo Nacional de Investigaciones Científicas y Técnicas’ / ‘National Council for Scientific and Technical Research’ (CONICET), the ‘Facultad de Humanidades / Faculty of Humanities’ (FH), ‘Universidad Nacional de Mar del Plata’ / ‘National University of Mar del Plata’ (UNMdP), Argentina. The research is developed within the ‘Instituto de Humanidades y Ciencias Sociales’ / Institute of Humanities and Social Sciences (INHUS, CONICET / FH - UNMdP) and the Research Group 'Ciencia, Tecnología, Universidad y Sociedad’ / ‘Science, Technology, University and Society' (CITEUS), FH – UNMdP, Argentina. [4] ONU - Organización de Naciones Unidas. (2023). Informe de los Objetivos de Desarrollo Sostenible: Edición Especial. https://unstats.un.org/sdgs/report/2023/The-SustainableDevelopment-Goals-Report-2023_Spanish.pdf [5] Vercelli, A. (2021). El extractivismo de grandes datos (personales) y las tensiones jurídicopolíticas y tecnológicas vinculadas al voto secreto. Revista Themis, N° 79, pps.: 111 – 125.

Lima: Revista Themis. https://revistas.pucp.edu.pe/index.php/themis/article/view/24867 [6] Vercelli, A. (2024). Regulaciones e inteligencias artificiales en Argentina. InMediaciones de la

Comunicación, 19(1), 105–135. https://doi.org/10.18861/ic.2024.19.1.3549 [7] United Nations. (2024, March 21). Seizing the opportunities of safe, secure and trustworthy artificial intelligence systems for sustainable development. https://digitallibrary.un.org/record/4043244/files/A_RES_78_265-EN.pdf [8] UNESCO (2021) Recomendación sobre la Ética de la Inteligencia Artificial. Paris. UNESCO.

https://unesdoc.unesco.org/ark:/48223/pf0000376713_spa [9] Library ONU. (2024, January 31). What does it mean when a decision is taken “by consensus”? What is the difference between "by consensus" and "unanimous"?. Dag Hammarskjöld Library. https://ask.un.org/faq/260981 [10] UN WebTV. (2024, 21 de marzo). Joint stakeout by US, Bahamas, Japan, Netherlands… [Video].

https://webtv.un.org/en/asset/k1a/k1a3snry9c [11] Vercelli, A. (2024). Libraries, access to information and artificial intelligences in Latin America and the Caribbean, en Garrido, M. & Wyber, S. (Editores) “Development and Access to Information”, pps.: 53 – 61, International Federation of Library Associations and Institutions (IFLA): The Hague. https://repository.ifla.org/bitstream/123456789/3334/2/da2i-2024-fullreport_rev-en.pdf [12] PIPT - Programa Información Para Todos. (2023). Plan estratégico del PIPT, 2023-2029. Paris:

UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000386173_spa [13] Maslej, N., Fattorini, L., Brynjolfsson, E., Etchemendy, J., Ligett, K., Lyons, T., Manyika, J., Ngo, H., Niebles, I., Parli, V., Shoham, Y., Wald, R., Clark, J. y Perrault, R. (2023). The AI Index 2023 Annual Report. Institute for Human-Centered AI. Stanford: Stanford University. https://aiindex.stanford.edu/wp-content/uploads/2023/04/HAI_AI-Index-Report_2023.pdf [14] United Nations. (2024, July 1). Enhancing international cooperation on capacity-building of artificial intelligence.https://digitallibrary.un.org/record/4054005/files/A_RES_78_311-EN.pdf [15] Boella, G., Mori, M. (2023). An Introduction to Ethics and AI. In: Chetouani, M., Dignum, V., Lukowicz, P., Sierra, C. (eds) Human-Centered Artificial Intelligence. ACAI 2021. Lecture Notes in Computer Science(), vol 13500. Springer, Cham. https://doi.org/10.1007/978-3-03124349-3_13

References

[1] AI Advisory Body . ( 2023 ). Governing AI for Humanity: Interim Report . United Nations. https://www.un.org/sites/un2.un.org/files/ai_advisory_body_interim_report.pdf

[2] AI Advisory Body . ( 2024 ). Governing AI for Humanity: Final Report . United Nations. https://www.un.org/sites/un2.un.org/files/governing_ai _for_humanity_final_report_en .pdf

[3] Vercelli , A. ( 2023 ). Las inteligencias artificiales y sus regulaciones: Pasos iniciales en Argentina, aspectos analíticos y defensa de los intereses nacionales . Revista de la Escuela del Cuerpo de Abogados y Abogadas del Estado, (9) , 195 - 217 . https://revistaecae.ptn.gob.ar/index.php/revistaecae/article/view/232