The First United Nations General Assembly Resolution on Artificial Intelligence – EJIL: Talk!

On Thursday, 21 March 2024, the United Nations General Assembly (UNGA) met to commemorate the international day on the elimination of racial discrimination. During that session it, almost clandestinely, adopted the first-ever resolution on the topic of Artificial Intelligence (AI). Towards the end of 2023, the UNGA already issued a series of resolutions dealing with technological development and their incorporation into the UN’s sustainable development goals. However, none of them dealt specifically and exclusively with AI. While the technology has been developing faster than ever, especially since the breakthrough launch of generative AI and Large Language Models such as Open AI’s ChatGPT, Anthropic’s Claude, Google’s Gemini or Microsoft’s Copilot in the past year, the progress on regulation has been creeping. The global nature of AI’s implementation requires an international approach to its regulation but it does not seem easy for States to find common ground. After years of debate, the European Parliament finally adopted the first regional framework, the AI Act, earlier this month, and the Council of Europe is currently working on a Convention on Artificial Intelligence. Still, on the international level, only little is happening.

This post will briefly explain the circumstances of the resolution’s adoption, and then shed some light on the differing interests of States regarding AI. Against this background, it will analyse the contents of Resolution 78/L.49.

Adoption of the resolution

The resolution’s draft was spearheaded by the United States (US). On 14 March it announced the submission of a draft resolution joined by 40 other member States with benefits “extend[ing] across the globe to countries at all levels of development”. The goals of the US were quite ambitious in that they stated from the beginning that the draft should aim at convincing all 193 member States of the UN. During her introductory speech, US ambassador Linda Thomas-Greenfield stressed that her country had been working with over 120 States to craft the text of the resolution. The final adoption happened within the UNGA’s 63^rd meeting of the 78^th session.

Although passed without a vote, over 125 countries decided to co-sponsor the resolution. Quite notable was an incident during the meeting when the Mexican delegate requested the floor to make a statement on behalf of Mexico, one of the State not co-sponsoring the resolution. However, the request came after the agenda item was closed, leading the UNGA’s president to deny it for procedural reasons. Interestingly, Mexico was the first Latin American State to announce a national AI strategy in 2018 and seems committed to advancing AI. It remains to be seen whether Mexico will still make an official statement on why it decided to refrain from co-sponsoring the resolution.

Highly diverging interests

This new resolution, while non-binding, is a first attempt to get all States and not just specific regions on board. Yet, interests regarding AI regulation are diverging extensively from continent to continent and even country to country.

The United States is home to the headquarters of at least four of the major players in AI: Open AI, Anthropic, Google, and Microsoft. Naturally, their interests align more with those of the businesses to foster development and revenue. While being aware of the risks for privacy and the dangers biased data encompasses, the White House on its Website with regards to a possible AI Bill of Rights maintains:

“These tools now drive important decisions across sectors, while data is helping to revolutionize global industries. Fueled by the power of American innovation, these tools hold the potential to redefine every part of our society and make life better for everyone.”

Conversely, Europe is known for its strict standards for data privacy. Both generative and surveillance AI require vast amounts of data to learn from patterns and produce sensible output. The European Union outlines a “European approach to trust in AI”, which is composed of “a European legal framework for AI that upholds fundamental rights and addresses safety risks specific to the AI systems”, “a civil liability framework”, and “a revision of sectoral safety legislation”. Accordingly, the European sphere focuses on reining in excessive data usage and protecting users.

The African States struggle mostly with accessibility and inclusion in AI advancements. Availability of AI is highly dependent on a connection to the internet. Of the 33% of the world’s population lacking internet access, the vast majority is located in the Global South. Moreover, the development, training and testing of AI relies heavily on data acquired from the Global North and is conducted mostly in English. This often results in data that is biased, discriminatory, and useless for the African market. Sandra Makumbirofa from the Research ICT Africa recently stated:

“The data that we have as African countries is not represented in the training of AI models. This means that the AI that we are using in Africa from foreign countries does not necessarily have the African context and therefore we are not able to use them efficiently as we can.”

States from the Global South rightfully fear being left behind in development and productive deployment of AI.

Contents of Resolution 78/L.49

It comes as no surprise that the common ground that could be found to ensure a wide acceptance of the draft resolution is relatively minor.

Still, the resolution appears more concrete and stronger worded than one would expect. It first and foremost reaffirms the commitment not only to the UN Charter but also to the Universal Declaration of Human Rights. Especially the reference to international human rights law is important for those States which propagate an ethical and safe use of AI in their national strategies.

Furthermore, two notions particularly stand out in Resolution 78/L.49 and presumably contributed to its success. The first one is a recurring reference to the “lifecycle” of Artificial Intelligence. This ensures a comprehensive approach that includes, on the one hand, the pre-training phase, the phase that is of particular importance for data privacy and selection, and, on the other hand, the technological development aspects and sale to the end consumer. Secondly, the text of the resolution repeatedly reaffirms that States must only fulfil their tasks in line with national law, ensuring a high discretion for States with regards to the implementation. The member States remain in charge of defining the necessities for their country and are not required to give up their national or regional policies, enabling them to continue to pursue individual interests.

Comparing the resolution to an earlier draft by the US of December 2023 provides some additional insights. While the draft “condemned” any misuse of AI, the final resolution only “encourages” member States to facilitate the development of frameworks to protect individuals from misuse amongst a number of other harmful practices. Further, member States were originally “called upon” to engage in several specific actions concerning AI governance. The final document reduced this to a sole “encouragement” for the States to act. In its last section the draft even “decided” to keep developments in the field of AI governance under consideration for relevant UN processes. The final resolution concludes with a mere acknowledgement of the UN’s role in reaching a global consensus on safe, secure, and trustworthy artificial intelligence systems. These changes show a softening in the resolution’s wording compared to the earlier draft, possibly significantly facilitating the acceptance of the document.

However, on one aspect the final resolution is, in fact, significantly stronger worded than the draft. Concerning the inclusion of developing countries, the draft initially only “encouraged” developed countries to assist other States to close the digital divide. In contrast, the final resolution “calls upon” all States to cooperate with and provide assistance to developing countries, stressing inclusive and equitable access to AI benefits and even naming particular actions that should be taken.

Conclusion

In summary, the first resolution on AI holds some promising conclusions, committing the member States to ensure the safe handling of the technology consistent with human rights obligations. Legally, it is certainly of a more symbolic character than that it actually carries weight. Nevertheless, it shows that States are aware of the complex problems that come with the development and deployment of AI and do not turn a blind eye on the dangers that the technology entails for their citizens. Not just the technological but also the legal developments in this new field are worth observing. This first resolution will likely be followed by many more that eventually ensure an ethical and fair use of AI, balancing human rights interests and different State interests.