WS #97 Interoperability of AI Governance: Scope and Mechanism

Summary

This discussion focused on the interoperability of AI governance, exploring its scope and potential mechanisms. Participants examined the concept of interoperability in AI governance, addressing issues that need global attention and obstacles to implementation. Key topics included the role of multilateral and multi-stakeholder approaches, the importance of trust-building, and the need for cultural and sustainable interoperability.

Speakers emphasized the significance of balancing regional variations with global approaches, highlighting the need for flexibility in governance frameworks. The discussion touched on the challenges of the digital divide and the importance of capacity building in developing countries. The role of the United Nations in global AI governance was a central point, with participants acknowledging its legitimacy while noting limitations in enforcement capabilities.

The conversation explored various forums and mechanisms for implementing AI governance, including the potential of blockchain technology and the need for agile responses to rapid technological advancements. Participants stressed the importance of streamlining efforts to avoid duplication and the need for clear mandates in international bodies.

The discussion also addressed concerns about AI sovereignty and data flows, suggesting potential solutions like “data embassies.” Speakers highlighted the need for a balance between efficiency and fairness in governance structures and the importance of prioritizing key issues in international forums.

Overall, the discussion underscored the complexity of achieving interoperability in AI governance and the need for continued dialogue and cooperation among diverse stakeholders to address global challenges while respecting regional and national interests.

Keypoints

Major discussion points:

– Understanding interoperability of AI governance, including legal, semantic, and technical layers

– The role of different actors (governments, private sector, civil society) in addressing AI interoperability

– Balancing regional variations with global approaches to AI governance

– The role of the United Nations and other international bodies in global AI governance

– Challenges of AI sovereignty and data flows between countries

The overall purpose of the discussion was to explore different perspectives on how to achieve interoperability in AI governance at a global level, while respecting regional and national differences. The panelists aimed to identify key issues, challenges, and potential solutions for creating more aligned and coordinated approaches to governing AI internationally.

The tone of the discussion was collaborative and constructive throughout. Panelists built on each other’s points and offered complementary perspectives. There was general agreement on the importance of interoperability and international cooperation, even while acknowledging challenges. The tone became slightly more urgent when discussing the need for the UN and other bodies to move quickly enough to keep pace with AI developments.

Speakers

– Olga Cavalli: Director of the South School of Internet Governance, Dean of the National Defense University of Argentina

– Yik Chan Chin: Associate Professor from Beijing Normal University, Co-leader of the IGF Policy Network on Artificial Intelligence

– Sam Daws: Senior Advisor of Oxford Martin Artificial Intelligence Governance Initiative, Director of Multilateral Artificial Intelligence

– Mauricio Gibson: Head of International Artificial Intelligence Policy, Artificial Intelligence Policy Directorate, Department of Science, Innovation and Technology, United Kingdom

– Xiao Zhang: Deputy Director of China Internet Network Information Center, Deputy Director of China IGF

– Poncelet Ileleji: CEO of Yoko Labs, Banjul, Gambia

– Neha Mishra: Assistant Professor from Geneva Graduate Institute, Switzerland

– Heramb Podar: Centre for AI and Digital Policy, India

Additional speakers:

– Dino Cataldo Dell’Accio: Chief Information Officer at the United Nations Venture Fund, Leader of the dynamic coalition on blockchain assurance standardization at IGF

Expanded Summary of AI Governance Interoperability Discussion

Introduction

This discussion, featuring experts from various fields and regions, focused on the interoperability of AI governance, exploring its scope, potential mechanisms, and challenges. The conversation aimed to identify key issues and potential solutions for creating more aligned and coordinated approaches to governing AI internationally while respecting regional and national differences.

Understanding AI Interoperability

The discussion began with a broad definition of AI interoperability, extending beyond technical aspects to include legal, semantic, and policy dimensions. It encompasses the ways through which different initiatives, including laws, regulations, policies, codes, and standards that regulate and govern artificial intelligence across the world, could work together more effectively and impactfully.

Speakers expanded on this concept, emphasising various aspects:

1. Yik Chan Chin stressed the importance of a broad definition beyond technical systems.

2. Sam Daws highlighted cultural interoperability and sustainability aspects.

3. Mauricio Gibson emphasised the need to balance regional variations with global approaches.

4. Xiao Zhang positioned AI interoperability as part of broader digital transformation.

5. Poncelet Ileleji underscored the importance of inclusivity and public interest in interoperability.

Key Issues and Obstacles for Global AI Governance

The discussion identified several critical challenges in achieving global AI governance interoperability:

1. Risk categorisation, liability, and training data risks (Yik Chan Chin)

2. Geopolitical tensions and unequal distribution of AI capabilities (Yik Chan Chin)

3. Sustainability and energy demands of AI systems (Sam Daws)

4. Keeping pace with rapid technological advancement (Mauricio Gibson)

5. Trust-building and data sovereignty concerns (Xiao Zhang)

Yik Chan Chin elaborated on specific global issues that need addressing, including:

– Harmonizing risk categorization across different jurisdictions

– Addressing liability issues for AI-caused harm

– Managing risks associated with training data and model outputs

– Tackling the unequal distribution of AI capabilities globally

Sam Daws particularly emphasised the need for an interoperable global approach to AI sustainability, noting the increasing energy demands of AI systems and the importance of measuring, tracking, and incentivising better energy and water use in data centres, chips, and algorithms.

Role of Different Actors in Addressing AI Interoperability

The speakers agreed on the importance of involving multiple stakeholders and diverse perspectives in addressing AI interoperability and governance. Key points included:

1. The importance of multi-stakeholder and interdisciplinary approaches (Yik Chan Chin)

2. The need for cross-regional forums for lesson sharing, such as the UN AI Advisory Body and regional initiatives like the Council of Europe and CAHAI (Sam Daws)

3. The government’s role in convening stakeholders and capacity building (Mauricio Gibson)

4. A multilateral orientation with multi-stakeholder engagement (Xiao Zhang)

5. The UN’s convening power and coordination role (Neha Mishra)

Mauricio Gibson emphasized the need for clarity in messaging and avoiding duplication in multilateral efforts. Xiao Zhang stressed the importance of finding priorities and focusing on them in UN efforts.

United Nations’ Role in Global AI Governance

The speakers generally agreed on the UN’s important role in AI governance, highlighting its legitimacy, capability for rapid response, and potential as a platform for dialogue and collaboration. Specific points included:

1. The UN as a platform for policy dialogue and collaboration (Yik Chan Chin)

2. The UN’s capability for rapid response in emergencies (Sam Daws)

3. The need to streamline UN agencies and define clear duties (Mauricio Gibson)

4. The UN’s legitimacy from equal representation of countries (Xiao Zhang)

5. The potential for a UN enforcement role in AI safety/security (Poncelet Ileleji)

Cultural and Inclusivity Aspects

An important area of consensus emerged around the importance of cultural aspects and inclusivity in AI interoperability. Sam Daws and Poncelet Ileleji both emphasised these points, highlighting the need for diverse cultural inputs into AI development, including insights from low-resource languages and indigenous peoples.

AI Sovereignty and Trust

In response to an online question, the discussion touched on AI sovereignty. Xiao Zhang emphasised trust as a fundamental issue in AI development and governance, shifting the focus from technical aspects to human and social factors. This led to further discussion on how to build trust in AI systems and the role of different stakeholders in this process.

Role of IGF and National and Regional Initiatives (NRIs)

Xiao Zhang highlighted the importance of the Internet Governance Forum (IGF) and its National and Regional Initiatives (NRIs) in AI governance. He emphasized their role in fostering multi-stakeholder dialogue and contributing to the development of AI governance frameworks at various levels.

Blockchain and AI Integration

Dino Cataldo Dell’Accio, an audience member, introduced the potential of integrating blockchain technology with AI to address interoperability and trust issues. He suggested this could provide a common layer of trust and transparency for AI systems, potentially enhancing interoperability and addressing some governance challenges.

Digital Economy Agreements

Neha Mishra brought up the relevance of digital economy agreements in relation to AI interoperability, suggesting that these agreements could play a role in facilitating cross-border data flows and AI governance.

Conclusion

The discussion underscored the complexity of achieving interoperability in AI governance and the need for continued dialogue and cooperation among diverse stakeholders. Key takeaways included the need for a broad understanding of AI interoperability, the importance of multi-stakeholder approaches, the challenge of balancing regional variations with global approaches, and the critical role of the UN in facilitating dialogue and coordination.

While there was general agreement on many points, unresolved issues remain, including how to effectively streamline and coordinate AI governance efforts across multiple UN agencies and forums, specific mechanisms to bridge the AI divide between developed and developing countries, and how to balance data sovereignty concerns with the need for global interoperability.

The Policy Network on AI (PNAI) announced its upcoming session and the release of its main report on AI interoperability and other issues. Participants were encouraged to attend the session and read the report for further insights on AI governance interoperability and good practices.

Olga Cavalli: Are we ready? Okay, thank you. Thank you very much for being with us at lunchtime. This is really remarkable. Thank you for being with us. My name is Olga Cavalli. I’m from Argentina. I’m the director of the South School of Internet Governance. By the way, our booth is over there. And I am the dean of the National Defense University of Argentina. And I’ve been invited to moderate this very important session. Thank you very much for inviting me as moderator. This is a great honor for me. This session is a workshop, Interoperability of Artificial Intelligence Governance, Scope and Mechanisms. So let me very briefly give the scope of this workshop. Interoperability is often understood as the ability of different systems to communicate and work seamlessly together. This is the concept that we all have about interoperability of different systems, software and machines. But the IGF Policy Network on Artificial Intelligence, also called PNI, definition of interoperability in the 2023 report is slightly different. I think it’s more broad, which is very interesting. The report, this definition includes the ways through which different initiatives, including laws, regulations, policies, codes, standards that regulate and govern artificial intelligence across the world, could work together in legal, semantic and technical layers that become more effective and impactful. This reminds me of the definition that was made by the WIEGEC about Internet Governance. That was a broader definition of Internet Governance, not only the technical identifiers and the technical coordination. At the same time, development and uptake of artificial intelligence systems are proliferating. them all the time at unprecedented pace and across sectors. A concerted effort in governing artificial intelligence is vital to harness the opportunities while remaining the challenge and risk as a result of new technologies. We all are working on different regulations in different countries and regions. As artificial intelligence is increasingly embedded in our society, it is critical that global governance frameworks encourage interoperability to promote a safe, secure, fair and innovative artificial intelligence ecosystem. So finally, interoperable systems and interoperable governance frameworks that effectively address the risks and impacts become really imperative. This is why we are here with a group of very distinguished panelists that I will introduce them now. We have Dr. Ying-Chang Ching. She’s an associate professor from Beijing Normal University from China and she’s also co-leader of the IGF Policy Network on Artificial Intelligence, PENI. Thank you for being with us, Ying. From remote, we have Mr. Poncelet Nyeleki. I hope I pronounced it correctly. He’s the CEO of Yoko Labs, Banjul from Gambia, Africa. Well, are you here? Hello. I can see you. We have Mr. Sam Doss. He’s senior advisor of Oxford Martin Artificial Intelligence Governance Initiative, Oxford University, United Kingdom. And also, he’s the director of Multilateral Artificial Intelligence. And we have Dr. Xiao Chang. She’s deputy director of China Internet Network Information Center, CINIC, and deputy director of China IGF. And we have Mr. Mauricio Gibson. Mauricio is head of International Artificial Intelligence Policy, Artificial Intelligence Policy Directorate. Department of Science, Innovation and Technology from the United Kingdom. So, welcome all of you. Thank you. And also, we have a very, very big audience, which is fantastic. And we have to deal with this noise and sound thing, but we will manage. Don’t worry. And also, we have a discussant that she will give us her input at the end of the interventions of our panelists, Dr. Neha Misha. She’s Assistant Professor from Geneva Graduate Institute, and she’s in Switzerland. So, we will organize our workshop in this way. We have three policy questions that will be answered by our distinguished panelists. And then, we have some comments from Neha from remote. And then, we will open the floor for the intervention of our audience. So, I will post first the first policy question to our distinguished panelists, which is about understanding the interoperability of AI governance. So, for the panelists, what is your understanding of interoperability? And what are the most issues that need to be addressed at the global level? And what are the obstacles? I don’t know who would like to start responding this question. I won’t put you on the spot. Okay. Welcome. The floor is yours.

Yik Chan Chin: Thank you, Olga. So, I speak on behalf of the PNAI because I’m the co-leader of the subgroup on the interoperability of AI governance. So, from a PNAI’s point of view, as Olga mentioned, we take a broad understanding of the interoperability. So, we particularly look at legal, semantic, and technical layer of the interoperability because we identified that the most important layers in terms of interoperability. So, we also look at how the law, regulation, policy, code, standard across a different part of the world can work together and address those important problems at a global level and make it more effective and impactful. So in terms of a global issue we would recommend to address in the short term or the medium term, there are several of them. I think the most important is AI models risk categorization and the evaluations. I think most of the country will agree on that. But we have a different approach in terms of the categorized risk and also evaluate mechanism. The second one, we identify liability, the liability of the AI system. And the third one is the risk of the AI training data. So we know that the AI system depends on the data that used to train the AI. So the risk of the training data is the third issue we think is important to address globally. And the third one, the last one is a technical standardization, the alignment of the technical standard and the regulatory fragmentation and the divergence requirement. So these are the global issue we recommend to focus on. So what is a major obstacle? The major obstacle, first of all, we identify is geopolitical tensions, the tensions between different powers in the world. And the second one is lack of the trust, lack of the trust among different countries, regions, and even the countries. And the third one is unequal distribution of AI technology and the maturity of their policymaking. So we see different AI technology power that have different power dynamics and the maturity of their governance. And the third one, that is about AI interoperability policy. Because we see a lot of the national, regional, international interoperability policy, but they have a different principle, values, objectives, and the priority. I think that’s it. That’s all for me. Thank you.

Olga Cavalli: Thank you very much. That was very interesting, especially following the comments in the Open Ceremony about the difference between the Global South and the North. Sam, you want to also tell us about what is your understanding about interoperability?

Sam Daws: Thank you very much. It’s a real pleasure and privilege to be here. I wanted to commend Yik-Chan Chin and the others for a wonderful P&I report on interoperability. Building on her comments, I would add that two areas I think that we need to additionally focus on. One is we need an interoperable global approach to the sustainability of AI. So AI and energy demands are set to grow with increasing multimodal inference with the use of IoT data and with agentic AI. So we need interoperable ways to measure, to track and to incentivise better energy and water use of data centres of chips, algorithmic efficiency and data sobriety. So work is already underway on this in the ITU, ISO, IEC, IEEE-SA or these Geneva-based acronyms. But also with the International Energy Agency on the energy track, partners such as the Green Software Foundation and the UN Environmental Programme, which takes a triple planetary crisis approach mapping the full life cycle of AI, all the way from mining through to end-of-life reuse. We also need international scientific collaboration for AI’s positive climate contributions. For example, in new materials research in solar PV and batteries and climate weather modelling through digital twins. also especially I think looking towards Belen in Brazil, COP30 next year, AI will really help deliver efficiency targets across all industries. So what are the obstacles to this particular issue area? Well building on Yip Chan’s remarks, national security and economic competition factors are significant. We’ve seen US export controls on high-end chips on China and in return China restricting gallium germanium antimony in response. Countries then race to acquire high-end chips detracting focus from building interoperability on sustainability approaches. The other obstacle is that tracking energy use by grids and companies can be economically sensitive. So companies don’t always volunteer this themselves and while companies have been doing a remarkable job, NVIDIA, Google and so on, in achieving 100x efficiencies in data centers and chips and software design, the overall electricity of use of AI continues to rise. So we need a multi-stakeholder framework for industry transparency and accountability. Singapore is a great member state example of integrating sustainability into its AI verify and model gen AI frameworks. And lastly I want to touch briefly just in one minute on cultural interoperability because it’s not talked about enough in AI governance but we really need cultural interoperability addressed at a global level. For humanity to flourish it’s vital that our diverse cultures feed into AI so we can better use it to live good and meaningful lives and that includes insights from low resource languages and also the wisdom of indigenous as people who have a minimal digital footprint, not captured by large language models trained on the internet. The trend for valid reasons towards sovereign AI at a national and regional level, especially in data governance and LLM worldviews, is, I think, going to continue. That’s not in itself an obstacle to interoperability. The obstacle would be if we have a fragmentation into a closed loop of culturally informed, epistemological, generative AI ecosystems – it’s a bit of a mouthful – such as a more socially conservative BRICS AI alliance that President Putin announced this week, an ecosystem around that, alongside a more liberal Western one, creating two bounded rationalities, separated by mistrust.

Olga Cavalli: About sustainability and how that affects our environment. Because we use technology, but technology has impact in the environment. That’s very interesting. Also, what you mentioned about the society and languages. Languages are my hobby, so I commend you for that comment. And I commend you both for being so respectful about the time, responding to the questions. Any comments from Echan or Mauricio? Yes, please, what’s for you interoperability?

Mauricio Gibson: People hear me? Yes. Thank you all for having me. It’s a pleasure to be here. I’m going to build on what everyone said, and it’s really helpful insights here, but give it a more practical government application perspective. So I think recognizing what people have said, I believe that, you know, there are innately going to be those different government interests, which at times will compete, but I think seeing how interoperability can happen is looking at the sort of broader areas where there are opportunities for cooperation and recognizing and honing in those particular areas and also looking at how we can plug the gaps and continue to build on those areas where there are gaps and sort of work towards that further progression of coordination. And I think a lot of that is sort of building those foundations and building blocks of what are the core principles that we’re starting to see across different governance work streams. And I think, you know, that doesn’t necessarily mean to be harmonization, but just really like building on that, because there will be the regional and domestic variation. And just, yeah, working on that gradually is a fundamental element of it. Though, I think what are the sort of core issues for interoperability and the obstacles that need to be overcome? I think, yeah, echoing what everyone said here, but also I think thinking more broadly about the sort of technological advancement of AI. And, you know, we’re hearing a lot more about not just gen AI, but agentic AI and the many challenges on governance of this, not least because of the nature of who is responsible building on liability there. You know, keeping up in terms of governance with these challenges is going to be a real battle. And so from the UK perspective, I think the science behind the most advanced AI, which is progressing, you know, at an exponential rate is a real focus, or has been a real focus, not least with the safety reporting that we’ve been producing, or secretariating the producing and bringing a lot of scientific evidence on this. And the state of the science, you know, it’s rapidly evolving, and we’re having to, you know, been having to produce a lot of reporting on a regular basis. And even by that point, you know, is that gonna be out of date? You know, how can we keep up with that? And understanding the scientific basis is gonna be a vital thing to try and overcome. I think the other thing is building on what people have said as a sort of capacity building element to, you know, I think there are differing understandings in different environments. The digital divide is so significant and given the advances in technology, we, you know, supporting policy officials, civil servants, public sector, and all, you know, everyone to support their AI talent uptake in all sorts of parts of the world will help an understanding of how they can engage in the governance process at the international level and in their own domestic system too. And I think a third point is supporting that is the sort of clarity and messaging that is needed to different communities across the world. So to support on things like sustainability or on the sort of cultural cross exchange of information, I think, you know, how do we land the key points that are needed to support that interoperability? And one element of that I think is using the different forums and I think what we’re seeing though, however, is in the multilateral domain, there is still a lot of duplication and the messaging isn’t clear and it’s not very clear how or where people want to sort of prioritize particular engagement on governance in these different areas. So some people are seeing some things happening in the UN, some people want to see in other areas, but one obstacle that needs to be overcome is the duplication, some of the activity and how can we try and manage that, see how they fit together and that’s gonna be a real challenge, I think, and something that we need to sort of work together on going forward.

Olga Cavalli: Thank you, Mauricio. You bring a very interesting point. It’s especially for developing economies. This is sometimes it’s hard to follow up with all the spaces where all these things are debated and decisions are made. made and policies are developed, maybe at the regional or global level, so that’s extremely challenging. And also you brought an interesting concept about capacity building, which is, as we know, cyber security, cyber defence, cyber crime, artificial intelligence, we are running short with the people trained. Yi-Chan, you want to add any comment?

Xiao Zhang: Actually, I think you might, I want to respond to your questions and I think you would respond to all the speakers. And we have three questions, one by one. The first one is your understanding, my understanding of what is interoperability of AI system. And for me, I think I use one word is one, it should be one ecosystem. Let me give example, I want to make comparison to internet. You know, for the internet in the past 50 years, only 50 years, and as I can say, it’s a one world, one internet. Why? It’s digital economy, it’s flourishing, there’s so many applications, but we have found something in common, and we divide internet governance to layers, at least in the technical layer, I mean, logic layer, we found TCP and IP, and we can connect them, we obey the same rules. That means, even though the contents, different countries have different regulations for the contents or something like that, but under the technical layer, we obey the same regulation. So that means we can work together, ecosystem. So as an internet user, you can use any application, you send email, you call VIP, or you search online, or something, you don’t feel that you are roaming around the world. seamlessly. I mean that we’re very very seamlessly. That means you feel very, you don’t feel that where it is. So I think for AI system, at least we should find something that we can work together as one ecosystem. So that’s my response to the first question. And the second one is what’s the priority, what’s the most important thing that we should do? I think actually because we have different culture, we have different development stages, and the priority for each country is not the same. For economic growth, for different areas. So our understanding of AI governance is quite different. I think because 2.6 billion people around the world have no access to internet. So AI means nothing to them. We cannot leave them behind. So maybe I think the most important thing is to sit down and find the priority of all the questions of AI and find priorities. What is the, we can narrow down, what is the global issue and what is maybe for developing countries and for Africa and for, we can one by one go. It’s not just AI risk or something like that. They have no AI. How to say AI risk? So I think development issue is also very important. And for the third one, so question is what’s the obstacle? As Professor Chen said, we actually have a lot of discussion. I think trust is the most important issue. There is, AI is built on trust. And it’s not limited to the geopolitical reasons. We shouldn’t have different ecosystem. So it’s all this ecosystem are built on trust. So how to build trust? I think this is something we need to discuss. Thank you.

Olga Cavalli: Thank you very much about trust. as we know, artificial intelligence is based off a big amount of data, capacity of processing that data, and some algorithms that gather that information. So trust is, I would say that it’s a layer overall that gives this tool the confidence for us to use it. So that’s a very interesting point from you. So I will address our distinguished panelists the second question, which is, how can different actors address interoperability and how can we balance regional variations with global approaches? Who would like to start? I don’t want to put someone on the spot. Xiao, Yichen, okay.

Yik Chan Chin: So yeah, I’m Yichen and this is Xiao. So yeah, I think I just jump in because from the PNAS perspective, just want to mention that we just released, now we’re going to release our main report in the main hall in the 4, 3.30. We have main sections. So in that main session, we’re going to report, release our report, this year’s report. So part of the report is interoperability and also liability and environmental issues and the labor issues. So welcome to join us in the main hall. So from the PNAS perspective, I think we are look at, I think how can we work together? I think multistakeholder is very important and the way from our own experience, cause I’m leading the group. So we got a lot of the input from different sectors and from around the world. So it just surprised how much information evidence you can collect through the multistakeholder models because we have a different sector, private government and academia. So that is really, really impressive for me personally. and as a group as well. And the second thing I think that interdisciplinary research is very important because it’s really complicated to understand the AI system and how to evaluate, how to test, how to know the security and the safety issue. So for us, I think it’s a multi-stakeholder plus the interdisciplinary research team is important. So in terms of how to balance between the regional variation and the need for global harmonization alignments, I think that’s a crucial issue because we have to respect the regional and the national diversity, where at the same times, you know, try to align at the global level. What we suggest is that, first of all, they have like a, do not recognize, we do not think actually we need only one global layer, like we have UN, of course, at the global level, but UN’s role, it’s not to everything. Actually, it’s more about the coordination. So we actually, we respect the regional diversity and the national diversity. So first of all, I think that we have to make sure local needs has to be met, just like Xiao mentioned. So we have local needs and regional needs. Then what’s happening at the next level, so it’s bottom up from a community, from national. Now we have a regional initiative, we already saw so many regional initiatives, like Latin America, African unions, and Asians, and of course, EU. So we have seen all these regional initiatives, but what we need to do in the end is how do we coordinate and meet all these national, regional initiatives to the global level. So that’s what we need to do in terms of interoperability. So definitely we can do this. The first way we can use that is in our report, we identified some existing very effective mechanism to do the global interoperability. For example, like we talk about UN as a multistakeholder platform for us to negotiate. and communicate. So this is one way we can have some policy dialogue. The second one, we can think about the international collaboration in terms of the AI safety governance. So we have a good example set up by the K government in terms of the test and the verified AI safety. And then there’s many AI safety institutes already set up in Europe and also in Japan, in US, even in China, they’re going to, we have a regional one, we do not have a national one. So this is a good kind of a collaboration. And the third one is kind of a technical industry self-regulation and the technical integration. So this is an existing mechanism already there, it can help to do the global interoperability. And then the second mechanism we can use is a compatibility mechanism. So for example, we’re talking about like a mutual recognition so we can have a mutual recognition with regulatory approach. So we have a divergence in terms of regulation, but then we can have some mutual recognition and all. So this is one kind of a mechanism. The second one, we can rely on the international standards set up by the IEEE ISO. And they also collaborated together, all these international standards setting under the ITU, they collaborated together as well to align with each other. Second, we can talk about a certificate, security certificate, okay. And also joint AI safety testing or alignment mandate. In the end, still we can have a harmonization of the AI regulation or even the harmonization of the AI principle and terminology. The last one I want to mention is very important in terms of national and regional policymaking. So the policymaker in the national domestic level and the regional level, when they do the policy, they should try to incorporate. international standard in their policymaking. So when they’re doing policymaking, first of all, of course, we have to respect a domestic public interest objectives. But at the same time, if they try to align with the regional international standard, which will reduce the unnecessary barriers and the cost in the end, you know, for the interoperabilities. So try to ensure alignments of global standard. And the last one is increasing international regulatory coercions. So reduce unnecessary sections and divergence about the regulatory framework. I think I’ll stop there.

Olga Cavalli: Thank you. Thank you very much. And also this difference between mainly developing economies by technology developed by developed economies. So, and once they have to develop the regulations, then they have to have that in mind. So thank you very much for your comments. Who would like to follow Mauricio? Sam, please go ahead.

Sam Daws: Building on Nick Chan’s comments, the regional approaches reflecting diverse cultural approaches and interests is not in itself a bad thing. It’s inevitable. And I think it can be very positive. But policy interoperability can be more difficult once nations have enshrined their approaches in law when negotiated regional agreements. At that point, tools like international crosswalks, other tools remain valuable to determine docking points and to clarify taxonomy and language differences. But in the future, I think we can do better in two ways. And I’ve tried a bit creatively about this question. First is to start earlier. So start early in understanding the nascent approaches of other regions at the same time that we draft our own national and regional approaches. Those of you familiar with UN negotiations, negotiating UN resolutions, know that once a region has negotiated a common position, it’s very hard to unpick it in the face of criticism or objections from other groups. So sometimes just being aware of the key concerns of other groups can allow subtle changes in language or in framing rather than on substance as we elaborate our own position, which then aids interoperability of approaches later. And I think we can consciously use the four tracks coming out of the UN Digital Compact and the H-Lab so that we can use them for early iterative knowledge. exchange through policy dialogue, through standards exchange, through scientific convening and through capacity building. The other sort of area I thought we could be creative with is using cross regional forums. So forums that have at least one member state from more than one region for lesson learning to reduce regional siloing of AI approaches. So let’s use cross regional political, cultural, economic and scientific forum at both a member state and a multi-stakeholder level. So we’ve got the IGF of course now, other examples, the Organization of Islamic Cooperation, the DCO that Saudi Arabia leads, the Digital Forum of Small States, Singapore leads, CISA in Central Asia that Kazakhstan leads, the Arab League, the Shanghai Cooperation Organization, the CIS, the Organization of Turkic States, OECD GPAY, BRICS and the Belt and Road. All of these have a contribution. And also we mustn’t forget the role of the International Science Council, the ISC and National Academies. I think those are vital, especially since two or three years ago, the ISC embraced the social sciences as well as the natural sciences. And I feel strongly that psychologists, economists and social anthropologists have important insights into how human behavior can be an obstacle to policy interoperability approaches. So we need them at the table. And then lastly, the network of AI safety institutes can also play a potential cross regional interoperability role. But I would say only if it can broaden its membership and its agenda to widen its relevance to the global South.

Olga Cavalli: Thank you, Sam. And thank you for. naming the examples of inter-regional spaces of debate, because I was going to ask you, but you already mentioned that. I will think about something for Latin America, or maybe we can talk about that.

Sam Daws: I would say CEPI-ELAPE for Latin America.

Olga Cavalli: Okay, thank you for that. And who would like to follow? Please, Yichen.

Xiao Zhang: Well, I’m Shell. I can add something. Well, definitely multi-stakeholder engagement is very important, but definitely I think the AI governance should be multilateral oriented. So I think I’m a little different from the two, but we find something common. Both multi-stakeholder and multilateral engagement are very important, but AI governance is very different. So I still want to make a comparison with internet governance. You know, when internet just happens, something happens, it’s not that it’s not harm, it do not harm, no harm normally, but AI is totally different. With the beginning of AI, we know it can bring risks. It could be like comparable to the atomic bomb, and we know it can, all of the life could be at risk. So it’s definitely because of the, because it could be using a weapon in the military, or because it needs the sanity or something like this. So it’s totally different from the internet. It must be the country oriented. So I think it’s a multilateral, very important things, the engagement on multilateral, they have resources. It’s not something technical problem, it’s something of the legal, it’s something understanding of AI, what it is, and it’s a harm it could bring. So normally, of course, multi-stakeholder engagement is very important, very important, but multilateral is definitely should, because they have the resources, we have the action to take. So I think the both of the start two sides are very important, and it’s totally different from internet governance. Thank you.

Olga Cavalli: I think you bring a very interesting- Very interesting point. So when you mean multilateral, you mean governments talking to governments? Is that the idea, like United Nations? And the interaction that you mentioned with the multi-stakeholder spaces, I think that would be the ideal way to work. Because governments have a special role in taking care of economy, security of the country, ordering the laws and all the environment. So very interesting point of view. Mauricio, you want to add something?

Mauricio Gibson: Thank you. Yeah, I mean, just building on what Chet was saying, I think, and what you were saying, Olga, about the role the government could play. I mean, giving that government perspective here, you know, we can convene a range of different stakeholders, you know, using that interaction, and engage these spaces to really understand the issues that are being reflected by different stakeholders, and help funnel that into action and policy domestically and internationally. I guess that’s a useful conduit that we can provide in delivering the needs of the stakeholders. Building on what you were saying, and others were saying, I guess, you know, there is the resource question. There is the resource that governments have. And I think building on, I think, what I was saying before, again, about capacity building, and the particular role governments can play by using that resource. So we can point to a UK-led AI for development program where we’ve invested almost 80 billion, 80 million, sorry, in development programs in Africa, now increasingly in Asia. And a lot of that can go into skills and to compute. And that’s a clear example of where we can really, like, leverage the resource that we have to support what’s going on the ground. You know, further action on sort of upskilling and governance is a key component of that too, not least, I think, in many areas. And I think particularly on safety, I think it’s an area that we’re trying to use our resource, our experience in convening. a safety institute using the AI Safety Summits to really highlight to a wider global audience about all the safety components and risks that have been mentioned by my colleagues here. I think a second point is also sort of the better communication of the key tools that support things like interoperability in the private sector. So practical examples I can give. So we’ve got the UK, as we funded the AI Standards Hub, which is an international networking mechanism that can help socialize technical standards across the world and bring together a different industry and a multi-stakeholder environment to really talk about these particular areas. And I think by having those conversations that can really bring to light a lot of the really areas that might come across as a bit difficult to access in the sort of standard setting community to a wider audience. We’ve developed sort of AI management essentials as well, which is a self-assessment tool to make sure that, you know, if you’re a business, you can support assurance, support the trustworthiness, developing things in line with policymaking, sorry, policy principles that might be of importance. So like transparency, accountability, things like this. But then, you know, thinking back to, I guess, the sort of public sector adoption element, how can we support and communicate to the public sector ourselves and really enhancing the processes for enabling them to really build the uptake on a lot of this too. And I think with that, you know, going a bit more deeply or a bit broader actually in terms of implementation. So obligations like, you know, we can talk about interoperability in terms of the, you know, the important principles that we might share, but it’s how do you help implement that in practice? And I think there’s a role for governments to support those mechanisms, working with regulators, ensuring that there is the necessary support, guidance and upskilling for those who are working domestically. to look at the international activity and bring it to the domestic level as well and translating the things that are happening at the international level, which we’re working together and doing that domestically. I think one particular example also in sort of the more advanced AI front is the work that the G7 has been doing on sort of the Hiroshima AI process, which is looking at codes of conduct for advanced AI. OECD is looking to implement that. And there’s a case of like monitoring and then keeping that going and a regular assessment of what’s going on to help implement obligations. And then I guess also like, how do we strengthen the foundational principles? So looking at what we were just talking about, so reinforcing whilst we’re implementing, it’s also important to like bring to light where those overlaps are with other areas. So for, I would give a practical example of recent engagements. So we’ve sponsored an OECD and African Union dialogue. The second one took place in Cairo a month ago. This was a really positive space where there was a workshopping on an African charter of trustworthy AI. So what was looked at was a range of different governance mechanisms and tools, including the OECD principles, which looks towards interoperability as well as the UNESCO ethics recommendations. Bringing all these together and looking at how we can draw on different things to support new work that is happening in the African environment. And we want to sort of continue with that work and help support it. So it’s kind of strengthening what’s out there, bringing those two things together and helping that communication and using the resource that we have to help support it is a really key thing. And finally, on your second part of your question about the sort of regional disparities and bringing that together in the global environment, how do we do that and getting the balance right? So the OECD African Union is a combination of two regional activities, bringing that together is a really helpful example. Another example is… This year we adopted and signed the world’s first AI treaty, which is the Council of Europe on Human Rights through Law and Democracy. This was really interesting because it brought together a global grouping and even with that, there were a few challenges in really like getting agreement on some of the core principles and the real detail, but we got there in the end because we were able to keep language broad and flexible enough to enable different global regulatory regimes to engage in it. And I think that’s the key thing. So obviously, whilst it’s a legally binding treaty, enabling space in text to support regional variation is gonna be really key in getting that balance. And I think that’s something that we have to sort of continue to recognize whilst we work towards interoperability and move and progress in these revolution of discussions in this space.

Olga Cavalli: Thank you, Mauricio, for this very good examples of cooperation. And I love the standards hub. I like very much that concept. You know, all the internet is global, it’s based on standards as you were mentioned at the beginning. So I think that agreeing on global standards is the key to have a global understanding of anything. So thank you very much, Mauricio. And now I will share the third question. And thank you all of you for being so respectful of times. Some of you took a little bit more, but some others a little bit less. So it’s a good balance in between all of you. And so the last question for you is, the role of the United Nations in global artificial intelligence governance. What role should the United Nations play in tackling the artificial intelligence governance? Who would like to start? Sam, please go ahead.

Sam Daws: Thank you very much. Before that, I’ve just noticed that Poncele from Gambia has disappeared from our screen and it would be valuable to get an African perspective. You’re there, great. so yeah make sure that he’s still accessible, great to see you. First and foremost the UN can help build trust to gender interoperability so this is very much building on Xiao’s point and trust is not a fixed constant it’s based on regular interaction so people-to-people contact which is why IGF is of such value it’s based on attitude we need to approach this issue with empathy with approaching knowing the other with curiosity and trust is built on experience so a track record of cooperation through predictable tracks so we can begin by an integrated global implementation of the two UN General Assembly resolutions agreed by consensus this year one proposed on responsible AI by the US co-sponsored by China and the other proposed by China on AI capacity building co-sponsored by the United States and both of those are guided by again universally agreed UNESCO recommendations on ethical AI so I think that is our foundation. Then I would suggest we focus on AI capacity areas capacity building in areas where cooperation has already shown to be able to be advanced despite geopolitical headwinds so this is areas like food security, biodiversity, climate change, health emergency prevention, macroeconomic stabilization, counter-terrorism and crime and data for the implementation of the UN sustainable development goals. The GDC and the HLAB on AI have given us a good roadmap it’s clear that the role of the UN is not or at least not for now to regulate AI nor is it to enforce compliance but that may come over time but the UN Secretary General can provide moral leadership on the need for inclusive AI. ITU, other agencies, DESA, UNDP can help bridge the AI digital divide through capacity building. The UN can be a source of scientific insights and expert data to guide decision making and convene policy dialogue in a standard setting. So lastly, and this is again trying to be a bit creative, I think the UN should look at the success of common security in the peace and security domain and look at whether those organizations could also play a role. I think if existing common security organizations established to build trust in the peace and security and economic domains, they could collaborate also in areas where AI can support shared objectives and knowledge exchange. So I’ve got a different set of acronyms here to the cross-regional ones, but the OSCE, CISA, ASEAN, the African Union, the EU, the GCC, the OAS from Latin America, Caribbean community from Gulag, the Pacific Forum, these are all examples where they have shown demonstrated ability to build confidence building through diplomatic engagement, which could be applicable. And finally, we’ve seen the emergence in the peace and security AI space of some very good initiatives by the Netherlands and South Korea, co-sponsored more recently by countries like Switzerland, Kenya, and others on the responsible AI in the military realm. And those have been very good. They’ve included China, which is really important. We’ve also seen US-China bilateral successful consultations on not using AI in nuclear guidance systems. I think we need to move in a direction of travel to back to the UN for AI peace and security. So the Security Council being more seized of AI in peace and security, it’s already done some work, as well as the UN’s work on non-proliferation and disarmament in Geneva and Vienna. Thank you.

Olga Cavalli: Thanks to you, Sam. Thank you for summarizing what has been happening in the UN. and your suggestions about the future. That’s very interesting. Who would like to follow the comments from Sam? Yes, please, the floor is yours.

Yik Chan Chin: Thank you, Sam, because we know he’s an expert in terms of the UN. So thank you very much for your comment, very insightful comment. So I think from the PNAS perspective, as we mentioned before, so international collaboration, the UN’s two resolutions is one example how to do the international collaboration. And the second one is very important. Actually, the UN has a function and also has a legitimacy to form the common objective of governance, because this is from the UN General Assembly. And so, for example, like a safe, secure, trustworthy, artificial intelligence, which is agreed in the General Assembly, that’s the two resolutions. So this is the two functions. And the third one is about IJF. The UN should strengthen the governance for the non-stakeholder, especially like a kind of police dialogue, which can have a marginal system structure to facilitate the exchange and to understand your policy and legislation. The best practice, of course, of the country and the cultures, because it’s very important that black people come to IJF. Sometimes, some colleagues told me, it’s just a talking show, because we talk here all the time. So we do not have enforcement power, okay? It’s very important to understand each other, build up the personal connections and a dialogue. So actually, it’s the most important, the multistakeholder platform, and that you should use it as a platform to have a dialogue, to have a dialogue, to have a dialogue with each other. built up a global AI dialogue. But at the same time, it also strengthens AIG’s capacity. For example, in terms of financial support, medical support, and resources support. Of course, there’s other overlapping functions by different agents. That’s why I really want to give one example, which is my personal example. So I’m part of the OEWG. Many people know that. So it’s about ICT securities. So the OEWG is a binary process. The Singapore ambassador took over as a chair, and then he was on the OEWG. So what he did for me, he invited a multi-stakeholder, like other sector, like private, and also an NGO, to participate in the OEWG, and give them a policy consultation role. So he listened to me, or Joe, some private sector joined in this consultation, was asking them, you know, the OEWG process, would you really help? Because I was invited as a representative from China, but I’m not, and to brief the delegate about the security issues. So we can see this is, the UN is doing some reform in terms of how do they cooperate with the multi-stakeholder dynamics and the multilateral process. So this is a very positive progress. So I’m going to the OEWG, and set up a presence, you know, other agents, other multilateral process. Thank you.

Olga Cavalli: Thank you very much. We already mentioned this, this side of the plan about multilateral and multi-stakeholder spaces. Ms. Yunxiao. Okay, sounds good. Yep.

Mauricio Gibson: Yeah, I’ll be right back. Thanks for the floor and yeah thanks my colleagues for setting that out. I mean yeah I think there are a lot of really interesting opportunities at the UN and I think from the perspective it’s a real opportunity here I think with the conclusion of the dualistic compact that’s been mentioned and you know there’s an opportunity I think for us to really like capture the opportunities that are presented with the UN’s convening power bringing every country and a range of stakeholders together through environments like this to really highlight potential of cross-cultural information exchange and sharing and building that mutual understanding and I think really highlighting and reinforcing the points that you were saying Chan about the building that understanding is I think really fundamental to the value add here in the UN and I think the one thing to sort of clarify is there are because of that fact that there are so many different UN bodies and agencies I think it’s really important to reinforce the importance of complementarity coordination role for the UN to to not duplicate but highlight where it has those different value adds depending on each relevant agency and an activity that is going on you know see as mentioned UNDP has a role in capacity building more widely on AI governance we’re seeing more interest across different agencies to play to play a bigger role but I think what we need is coordination so and an understanding through each agency what exactly is needed to be delivered on the ground giving that practical benefit moving away from just conversation about the principles on interoperability and coordination between these but actually supporting coordination on the ground and delivering actual benefits to those communities who are most acutely feeling the digital divide as well. I think just one of the ways of actually delivering on that, you know, looking at what you mentioned, Sam, about these global dialogue on AI governments, such as one initiative, which has been proposed through the final Global Digital Compact text, and which is just about to launch in terms of negotiations for the modalities of it. I think it’s important that we really, like, highlight the opportunity for sharing information in these forums and actually building that understanding somewhere like this, highlighting the different initiatives that are around enforcing those points about, you know, these are the actions that we are doing, how to build and understand that, bringing those together. And there is a role, you know, for the IGF to be considered, and it’s interesting you mentioned that, and the role of the OEWG, you know, kind of, you know, we need to, like, consider these in these next stages of thinking about that. And I think, on top of that, it’s important that we don’t, you know, create too many things. It’s meant to be in the margins of existing conferences. How can we leverage those existing activities? You know, the ITU AI for Good Summit, there is the UNESCO Global Forum on Ethics as well. How can we work together on these? And then you also mentioned the sort of scientific panel on AI, and so the UK is very interested in this because we produced safety reporting, so on advanced AI risk, which is on an expert panel, leading the scientists, synthesizing research, and I think there’s a role for this. And the role for the UN to really highlight and reinforce a lot of the research that is out there and bring it to a wider audience so we can support that inclusivity in terms of the understanding of science and actually move forward when you have that understanding of science. But I think, again, like, you know, it’s ensuring these are clear and grounded in, you know, scope, the scope is clear, the mandates are clear, so we don’t get into a situation where things are muddied. And that is also reflective of why a lot of people. We are talking about the WSIS process. There are consideration of the role of AI in this process, but we need to make sure that there’s effective coordination so it actually best delivers for people through effective AI governance. And just the final thing to underpin a lot of this is, as I mentioned before, there’s differing approaches, there’s technological advancement that is moving so quickly. It’s vital that we stay alert to the need for agility in AI governance and flexible approaches that can adapt to different developments in the world. I think at times, the UN might not be the quickest moving institution and the system is not the most quick moving, but we need to recognize that we need to keep up with the advanced technology and that’s a fundamental thing as well.

Olga Cavalli: Thank you, Mauricio. I like the concept that you mentioned about the United Nations being the point of… Spreading information to other countries that are part of it. Yixuan, the floor is yours.

Xiao Zhang: Absolutely. As Chris said, I totally agree. At the UN, the United Nations, the best place… Governance… Park… Is… Should actually be more important as it is based on… Yixuan was telling me… Some successful experience in… Like… Climate… Sensitivity… And… So, I think… Something totally different… It’s actually part of the digital transformation of economy and society. It should not be something very… It’s not a single thing. It should be… Actually, it’s part of the digital… Part of the economy, society, and system. So, I see this provides a best place for us to… And, you know, I can come here and I don’t have the energy, resources, and budget to go anywhere. So, I can come here once a year, but I do have, for example… Also, in the UN system, not only… Which is here, but also you have… Again, it’s a must-see program. And I just think it’s a good one. So, we need simultaneous… I think… Resources, and we have… You know, now we have to… I’m not sure if I’m… You know, it’s just like… It’s important to keep making up, but like… And this should continue. I’ve seen how the GDC has something… Something… Solid tasks, solid tasks. And I believe that IGF could… To carry… To carry some of the tasks. That should be the main track. Thank you.

Olga Cavalli: Thank you very much. And for this sort of combination of United Nations and IGF. Yeah, interesting. We have not forgotten you. I will give the floor to you now that our distinguished speakers here in the room have already answered the questions. Would you like to comment about the three questions that we have been talking about? What is the role of the United Nations and how different actors could interact to work on this very important issue? Welcome, the floor is yours.

Poncelet Ileleji: Thank you very much, Olga. Thank you to all the speakers, Professor Chin, Sam and Mauricio. I would like to say first and foremost that we have seen all the three questions that were asked. All my colleagues and speakers, talking from a PN AI perspective, spoke about the three key pillars of what we are talking about in terms of measures, tools and mechanisms, in terms of interaction and interconnection and in terms of communication and cooperation. And I will want to focus, coming from the Global South perspective, I would like to focus on the communication and cooperation part, you know, I have to be a little bit biased here. And I’ll say one thing that guides me in this is for us to remember that at the end of the day, we have in September the Governing for AI for Humanity by the UN advisory body. And one of the key recommendations was about the set up of an independent international scientific panel on AI, which should be multidisciplinary. And we also have issues that I feel that… Some talked about them, and one that was very key for me, if I relate it to recommendation one of that UN advisory body report, deals with producing quarterly thematic research, which will help achieve the SDGs. And when we look at whether it’s climatic issues, when we look at poverty, these are things that AI can be used as an enabler. We have to remember that at the end of the day, we want people to have inclusivity. We want public interest to be well-represented. And within the policy network for AI, we try to look at things from that perspective. In as much as possible, we have various stakeholders, but we try to look at the constituencies we come from. And that is why aligning it with all the regional initiatives, whether it’s the African Union, whether it’s the EU, is very important. But I think if AI can make a difference to us achieving the SDGs, we have gone a long way, where we build on trust and equity. Thank you.

Olga Cavalli: Thank you very much for that, especially as I think about trust and contributions about countries and organizations. I would like to give the floor now to Miha. She has been patiently listening to all of what our colleagues have been saying. Miha, what’s your comments about the debate and exchange of ideas that we have been having?

Neha Mishra: Thank you. Thank you very much, Olga. And also I joined the others in congratulating the PNI for the report, and I’m so delighted to be a part of this panel. So the discussion has been an embarrassment of riches. I really don’t know how I can add, but I wanted to weave some of the ideas that I thought was common through the discussions. The first thing I thought that was very interesting… is the different dimensions of interoperability that the different speakers mentioned. In addition to the technical, legal, and the semantic interoperability, which is often discussed, there were also other dimensions of cultural interoperability or sustainable or environmental-related issues being brought together. I think it was quite interesting when some of the governmental perspectives were shared, particularly how to navigate the different interests of different governments to figure out an interoperability framework that might be feasible. Also, I think here from a practical implementation perspective, questions might be relevant in terms of thinking about whether it needs a more modular approach, whether it’s something to be tested in specific sectors, how incremental it should be, and what the prospects of a multi-stakeholder approach are, because one thing that I thought was common through the discussion also was that multi-stakeholderism and multilateralism need to align with each other, and there can be certain tension points that need to be resolved. I also found it very interesting that a lot of the speakers, including Poncelot, brought this idea of the developmental divide, the AI divide, and there was a lot of very, very encouraging discussion on how to bridge the different gaps. I think one perspective I would like to add to that is that while it is great to think of capacity-building initiatives of more meaningful international regulatory cooperation, one should also be conscious about the limits of interoperability in the sense that in certain scenarios, developing countries, least-developed countries may not be able to participate in many of the interoperability dialogues. So to that extent, it is important to assess what are the areas in which we are looking for interoperability and how representative those discussions are. And while I fully encourage that it is important to have these open dialogues, to have more sustained technical and capacity building initiatives, this is an incremental slow process and developing countries should not lose their autonomy to decide how they want to develop their AI frameworks given that it can have very specific influences across different communities. And that’s why it was very important, I think, at the beginning to highlight the cultural aspect of the human layer of interoperability into the discussions. I also found it very interesting that we discussed so many different variety of tools and mechanisms and different stakeholders, different organizations, including at the global level, the UN, that can contribute to different aspects of interoperability. But at the same time, I agree with Mauricio that it’s important to streamline these efforts and to not duplicate these efforts. From a Global South perspective, I think the question is also really very practical as to if there are multiple fora, they only are able to invest that many resources and they might have to choose between different fora and that also can create competition between different fora. And in that sense, I think the UN still has a very continued important role as being the umbrella organization or the framework organization where a lot of, at least the high level values could develop. But at the same time, I think it’s inevitable and that’s why I think it was very helpful that Sam mentioned so many different examples, both of intra-regional, inter-regional and different kinds of even transnational policy networks. And I think even Mauricio mentioned how the private sector could be involved because I think between In setting these high-level principles and achieving them in practice, there are many, many different stakeholders, including private sector, civil society, academics, engineers, technical bodies, different cultural groups, different communities, and really bringing them together is not an easy task. So it was quite helpful to have that overarching perspective. I think one last point I would like to mention is that, and this is a question that we often think of, even from my disciplinary training as an international lawyer, one question we often think of is, you know, how multilateralism is changing in the current world, and even in the context of, I think, looking at AI interoperability, I think, especially because the development of the technology is not necessarily always state-driven, but also driven by a variety of private organization and standard development bodies, I think the need to find better modalities of engagement between the multi-stakeholder bodies, the transnational regulatory bodies, the private sector bodies, and the multilateral bodies is important. And I don’t think it is going to be a perfect process. I think it is about continuing efforts and figuring out what are the tension points, geopolitical conflicts that are completely not resolvable, and what can be, and it was great to see also many examples being discussed, where despite all the geopolitical differences, the developmental differences, there are common points of consensus and coordination that one can see at the UN level or at other international or regional bodies. I’ll end my comments here. Thank you so much.

Olga Cavalli: Thank you, Priya, for such a very concise and complete summary of what has been discuss and I like the concept that you mentioned that this is a process. I think the journey is a destination that we are going through. We have been talking about internet governance for almost 20 years so far, so maybe we think how many years are we going to talk about artificial intelligence. I would like to give the floor now to our nice audience that have been patiently waiting for an opportunity to talk. Please grab a mic and introduce yourself and tell us your name

Dino Cataldo Dell’Accio: and your organization. Thank you very much for the very engaging and enlightening discussion. My name is Dino Delacroix, I’m the Chief Information Officer at the United Nations Venture Fund and I’m also involved in the IGF. I play several roles. One of my roles is to lead the dynamic coalition on blockchain assurance standardization. The comment and question that I would like to pose to the speakers is I think it’s time to acknowledge that AI does not work in isolation but there is a convergence of AI with many other technologies and I think there is an opportunity for example see how the convergence between AI and blockchain can indeed address many of the issues presented by the interoperability needs. There were many references to trust and I think blockchain can indeed provide that common layer of trust in demonstrating that there is at least a data source because on blockchain we can store data sets that can be audited, can be verified in a transparent manner, can be validated as an input to the AI and also there is a synergy between the two technologies not only one way but both ways. The ability of AI to calculate and predict the volume of transactions for blockchain can usually it to be relatively slow and not performant, can help blockchain scalability issues. So that’s the point I wanted to bring to attention. Thank you.

Olga Cavalli: Thank you very much for your comments. Any other questions, comments from the audience? Do we have online questions?

Heramb Podar: Yes, so we also have a number of comments and a question from the online audience. So I think a bunch of you touched on how multi-stakeholder bodies such as the UN are in the point of information sharing, but they are ultimately relatively slow. So what would you want to see in terms of improving them to address this kind of rapid pace of technology and to keep up with that pace?

Mauricio Gibson: Yeah, that’s a very good question and yeah, it touches on the last points I was making at the end there, which I think is a fundamental question. I think the challenge is, if we’re getting to the crux of it, so UN reform is a long drawn out process. I think we’d have to think about these new stages of AI governance as we move to the next chapter of implementing the Global Digital Compact and WSIS and consider what are our core priorities we need to work on to achieve that agility. For example, with the scientific panel, we need to learn lessons and draw on the experiences of previous scientific panels that have been developed. Some have taken a lot longer than others. Some have different parallel political negotiating processes. I think if there’s a way of connecting with sort of more multi-stakeholder ad hoc engagement, so for example, the UK’s international AI safety reporting, you know, secretariat outside the UN, if we can draw on the experiences of existing initiatives that doesn’t require new UN bodies necessarily to be stood up to take a while to keep up with the technology, that would be a much better way of being more nimble and agile to the advances of the technology and that can apply in other areas as well.

Heramb Podar: So, another question we had from the online audience is how do we promote, you know, AI sovereignty? A lot of countries are worried about, you know, how data flows, how, who keeps the data and, you know, there’s a lot of countries, particularly India as an example, Russia as an example, who want to develop their own kind of national intelligences. So how do we balance that and balance, you know, this kind of like broader interoperability conversation around, you know, global south inclusion and having a unified approach?

Olga Cavalli: Who would like to take the question? Sam, go ahead, the floor is yours.

Sam Daws: So just on blockchain, absolutely, I think indelible, sorry, indelible ledgers are, I think, a real tool in increasing accountability in the future. Blockchain is one such thing, and it’s great to hear that someone involved in the UN pension fund is thinking about it for that purpose. On the speed of technology, we’ve got an exponential increase, and not just the UN, but I think governments are finding it very hard to actually set policies in response to it. The UN is capable of very rapid response. I worked in Kofi Annan’s office in the early 2000s, and we worked on a 24-7 response time for conflicts around the world. So if you look at the IAEA, the UN Security Council, the work of the World Food Programme in UNHCR in emergency situations, WHO in Ebola outbreaks and so on, we see remarkable ability to come to speed. But member states must want that capability. Member states have failed again and again at providing an independent preventive capability and strategic forecasting and these sorts of areas to international organisations. So I think there’s that, and where the UN takes a longer time, it’s often valuable to actually grow understanding of culture issues over time. So I think there is a role for the UN to be slow and steady, and there’s a role for the UN to be fast. I think it’s only after we have a major AI accident, God forbid, that we’re likely to see agreement that the UN can have the capacity for enforcement in the AI safety and security realm. In the meantime, the UN will rely on each nation state’s intelligence, military, foreign affairs, other resources to be able to monitor threats and challenges in real time. And lastly, on data sovereignty, there’s an interesting idea that’s floated of data embassies where you can have data of your own country stored somewhere else where perhaps you can have renewable power to power the data centre quite cheaply, but that that data is inviolable in the same way that diplomats are inviolable. So you can have these little kind of data embassies around the world. I think that’s an interesting concept that could be developed further.

Yik Chan Chin: OK, yeah, I think the blockchain, you know, I think in China or even in the private company already use the blockchain to do the security and encryption. So yeah, I agree with that. In terms of UN’s role, I think just like Mauritius said, there’s so many overlapping agents. So in order to streamline, they need to have a clear definition of each duties and reduce the overlapping. And I agree with Sam, because I’m personally involved in OEWG process. I think the UN has a big capacity, you know, collect and reach out to around the world, collect information from multi-stakeholder, even bilateral countries, you know. But the negotiation between states is so slow, personal experience, very slow. So how can they speed up the negotiation process? I think that’s the key. And in terms of sovereignty issue, AI sovereignty, my colleague published a paper on that. So I think the thing is, we need to figure out, as I said, what should be solved internationally, globally, what should be left for the individual country, left to their own jurisdiction. And this has to be discussed. This is a process, we have to reach agreement on that. Just like internet, you know. Some, like, we have a core infrastructure, which is public good, even global public good. But in terms of the content moderation, which is left to the national jurisdictions. So we need to have common jurisdictions, and at the same time, national jurisdictions. Thank you, that’s my comment.

Xiao Zhang: Yes. I want to respond to the online question of the UN, United Nations. Well, I think UN is not perfect, and a lot of limitations. But it’s better than nothing. And you see, I think always there is a balance between efficiency and fairness, and always a balance. So I think what we should do, maybe we call on the leadership in the AI era. Because the leadership’s awareness of what is happening is so important. And it should be AGL, assumption, engagement, something like this. And also, what I should suggest, is narrow it down. We find some priorities on the UN agenda. GDCL following, and find some priority, and we can focus. And step by step, I think the UN still could play a very important role in the AI era. And besides, I think for the engagement of the IGF and the multi-stakeholder approach. And I think, actually, we can strengthen this approach, because we have NRI, the National Regional Branch of each IGF. It’s very, very important. And if we have every NRI, it can support the policymaking of AI. So that’s my point. And I still think UN and IGF could play an important role in this era.

Olga Cavalli: Thank you all very much. Any other questions from the audience? From online? Haram? No. So I will give the floor, one last comment from. each of our distinguished panelists who would like to start, and then we have aquestion online.

Heramb Podar: Okay, can you read it? Yes, so IamSarmanova asks, in which specific forum do you see global implementation focused AI governance coordination taking place, especially as we think about, you know, duplication between all of these different forums and potential geopolitical tensions or baggage which might come with certain forums?

Olga Cavalli: Who would like to take that? I think, Sam, you’re the expert.

Sam Daws: I’m not the expert. Everybody is here. I mean, I would say that the UN being the treaty-based universal body is the go-to for where we can to implement AI governance mechanisms. So that’s the UN General Assembly, its six committees, and the tracks that the Global Digital Compact have set in train. And I think we’re going to see wonderful synergies within and across those four tracks going forward, and I hope we can then bring in all the valuable regional, mini-lateral, and national approaches into that. But the UN is only as strong as the willingness of its member states to cooperate. So the UN is great at the level of principles, but as I said, I don’t think there’s appetite among member states to get into regulation and enforcement. So we’ll need into operability, which I think is one of the purposes of this panel, is that reality. Thank you.

Olga Cavalli: Thank you, Sam.

Yik Chan Chin: I think I totally agree with Sam. I think the UN, one of the reasons, I think a very fundamental reason, is the UN, every country has one vote. So it’s equal, okay? Because no matter you are small, medium, or the strong nation, each country has one vote. So this gives the fundamental legitimacy of the UN, you know, we have an equal fit to participate. So yeah, so UN will, from PNN’s perspective, you can read our report. So we support UN as the focal point for AI governance or dialogue, but the enforcement, you know, it’s just like Sam said, how the state gives, each state gives the power to UN, you know, to at least laterally, in terms of the safety issue, security issue, maybe we can give more power to UN for the enforcement. Thank you.

Xiao Zhang: I totally agree with Ike.

Olga Cavalli: Thank you. Any comments from Mauricio Bonslet or Niha? Yeah, thank you very much, colleagues.

Poncelet Ileleji: I totally agree with all my colleagues, and I would encourage colleagues to read the SOP group report on AI governance in terms of interoperability and good practices that was led by my colleague, Professor Yeh Chin, who did a fantastic job on that. It covers a lot of stuff. And no matter what happens within interoperability, we have to know that public interest and inclusivity matters. And that will be my closing remark. Thank you very much.

Olga Cavalli: Thank you very much. Neha, any comments, final comments?

Neha Mishra: One thing we haven’t spoken about at all, but I just want to add it to the mix, is that increasingly, certain digital economy agreements are also looking at interoperability-related issues and trying to find synergy between technical and regulatory interoperability. And that’s something I want to add to the mix, because we haven’t discussed it at all. And I think there are prospects. I think there are prospects, especially at the regional level, or between like-minded countries that sign these digital economy agreements. Thank you.

Olga Cavalli: Thank you, Neha. Thank you, Poncelet. We have four minutes. I will give the floor for the last comment. We have more comments from audience? Any other questions from the audience, from online? No, from the audience. OK. Last comments of the three, four minutes that we have. I cannot read from five minutes. I know. OK. I think we have had a very interesting session. Thank you all very much. Thank you, I-Chan, Chiao, Sam, Mauricio, Neha, Haram, Poncelet, and all the audience. Thank you for being so patient and so active in participating in this very important session. Thank you for allowing me to moderate it. You want to say something? Please go ahead.

Xiao Zhang: I propose a group picture.

Olga Cavalli: Oh, a picture. Yes, that’s very important. Now we take a picture. And yes.

Heramb Podar: Just a very quick note. I’ll share the link to the interoperability report. I’ll share the link to the interoperability report. since it was mentioned by a lot of my speakers and a lot of the panelists also, co-authors on the report. We also have, for those in person, the PNAI session happening in the main Peter E. Hall, which starts in, I believe, about 20 minutes. So we look forward to seeing you there. And if you have more questions or you want to just know more about the PNAI work, please feel free to join. Thank you.

Olga Cavalli: Thank you for that. I participated and I was one of the leaders of the PNAI and labor issues. So thank you for allowing me also to do that. No sacas una foto? OK. Let’s do the picture. Thank you all very much. Thank you. Thank you. Thank you. Escuchame, como se puede borrar esa imagen? No, eso. Yes. Can you remove that?

Heramb Podar: I don’t. I can’t have that. They will have to remove it. Give me one second. Give me one minute.

Mauricio Gibson: Oh, no. We’ve lost him. Poncele, can you come back? Poncele? Poncele, can you come back? Poncele? Can you hear us? Poncele, are you there? Hello? OK.

Agreement Points

Importance of multi-stakeholder approach

Yik Chan Chin

Sam Daws

Mauricio Gibson

Importance of multi-stakeholder and interdisciplinary approaches

Cross-regional forums for lesson sharing

Government role in convening stakeholders and capacity building

The speakers agree on the importance of involving multiple stakeholders and diverse perspectives in addressing AI interoperability and governance.

UN’s role in AI governance

Yik Chan Chin

Sam Daws

Xiao Zhang

UN as platform for policy dialogue and collaboration

UN’s capability for rapid response in emergencies

UN’s legitimacy from equal representation of countries

The speakers agree on the UN’s important role in AI governance, highlighting its legitimacy, capability for rapid response, and potential as a platform for dialogue and collaboration.

Similar Viewpoints

Both speakers emphasize the need to address the challenges posed by the rapid advancement of AI technology, including its sustainability and energy demands.

Sam Daws

Mauricio Gibson

Sustainability and energy demands of AI systems

Keeping up with rapid technological advancement

Both speakers highlight the challenges posed by geopolitical tensions, unequal distribution of AI capabilities, and issues of trust and data sovereignty in AI governance.

Yik Chan Chin

Xiao Zhang

Geopolitical tensions and unequal distribution of AI capabilities

Trust and data sovereignty concerns

Unexpected Consensus

Cultural aspects of AI interoperability

Sam Daws

Poncelet Ileleji

Cultural interoperability and sustainability aspects

Inclusivity and public interest in interoperability

Despite their different backgrounds, both speakers emphasize the importance of cultural aspects and inclusivity in AI interoperability, which is an unexpected area of consensus given the often technical focus of AI discussions.

Overall Assessment

Summary

The main areas of agreement include the importance of multi-stakeholder approaches, the UN’s role in AI governance, the need to address rapid technological advancements, and the significance of cultural and inclusivity aspects in AI interoperability.

Consensus level

There is a moderate to high level of consensus among the speakers on key issues, suggesting a shared understanding of the challenges and potential solutions in AI governance and interoperability. This consensus implies that there is a strong foundation for developing collaborative approaches to AI governance, although differences in emphasis and specific concerns remain.

Different Viewpoints

Role of multilateralism vs multi-stakeholderism in AI governance

Xiao Zhang

Yik Chan Chin

Xiao Zhang argues that AI governance should be multilaterally oriented while recognizing the importance of multi-stakeholder engagement. She emphasizes that countries must lead AI governance due to its potential risks and impacts.

Yik Chan Chin emphasizes the importance of multi-stakeholder and interdisciplinary approaches in addressing AI interoperability. She suggests that these approaches can help collect diverse information and evidence from different sectors around the world.

While both speakers acknowledge the importance of multi-stakeholder engagement, Xiao Zhang emphasizes a stronger role for multilateral, country-led governance, while Yik Chan Chin places more emphasis on multi-stakeholder approaches.

Unexpected Differences

Overall Assessment

summary

The main areas of disagreement centered around the balance between multilateral and multi-stakeholder approaches in AI governance, and the specific ways to improve the UN’s effectiveness in this domain.

difference_level

The level of disagreement among the speakers was relatively low. Most speakers shared similar views on the importance of interoperability, the need for global cooperation, and the significant role of the UN in AI governance. The differences were mainly in emphasis and specific implementation strategies rather than fundamental disagreements. This suggests a generally aligned perspective on the topic, which could facilitate progress in developing global AI governance frameworks.

Partial Agreements

Both speakers agree on the importance of the UN’s role in AI governance, but they differ on how to improve its effectiveness. Sam Daws focuses on the UN’s potential for rapid response given member states’ support, while Mauricio Gibson emphasizes the need for streamlining and clear definition of duties among UN agencies.

Sam Daws

Mauricio Gibson

Sam Daws argues that the UN is capable of very rapid response when needed. He suggests that the UN’s speed in AI governance depends on member states’ willingness to provide it with the necessary capabilities.

Mauricio Gibson suggests the need to streamline UN agencies and clearly define their duties to reduce overlapping and duplication of efforts. He emphasizes the importance of coordination and understanding what each agency needs to deliver on the ground.

Similar Viewpoints

Both speakers emphasize the need to address the challenges posed by the rapid advancement of AI technology, including its sustainability and energy demands.

Sam Daws

Mauricio Gibson

Sustainability and energy demands of AI systems

Keeping up with rapid technological advancement

Both speakers highlight the challenges posed by geopolitical tensions, unequal distribution of AI capabilities, and issues of trust and data sovereignty in AI governance.

Yik Chan Chin

Xiao Zhang

Geopolitical tensions and unequal distribution of AI capabilities

Trust and data sovereignty concerns

Key Takeaways

AI interoperability needs to be understood broadly, encompassing technical, legal, semantic, cultural and sustainability aspects

Multi-stakeholder and interdisciplinary approaches are crucial for addressing AI interoperability challenges

There is a need to balance regional variations with global approaches to AI governance

The UN has an important role to play in AI governance, particularly in facilitating dialogue and coordination

Trust-building and addressing the AI divide between developed and developing countries are key challenges

Rapid technological advancement poses challenges for governance frameworks to keep pace

Resolutions and Action Items

The Policy Network on AI (PNAI) will release its main report on AI interoperability and other issues

Participants encouraged reading the PNAI report on AI governance interoperability and good practices

Unresolved Issues

How to effectively streamline and coordinate AI governance efforts across multiple UN agencies and forums

Specific mechanisms to bridge the AI divide between developed and developing countries

How to balance data sovereignty concerns with the need for global interoperability

Concrete steps to make UN processes more agile in responding to rapid AI advancements

Suggested Compromises

Using existing cross-regional forums to facilitate dialogue and lesson-sharing on AI governance

Leveraging both multilateral and multi-stakeholder approaches in a complementary manner

Focusing UN efforts on coordination and high-level principles rather than detailed regulation

Allowing for regional variations in AI governance approaches while working towards global alignment on key issues

Interoperability is often understood as the ability of different systems to communicate and work seamlessly together. But the IGF Policy Network on Artificial Intelligence definition of interoperability in the 2023 report is slightly different. The report, this definition includes the ways through which different initiatives, including laws, regulations, policies, codes, standards that regulate and govern artificial intelligence across the world, could work together in legal, semantic and technical layers that become more effective and impactful.

speaker

Olga Cavalli

reason

This comment introduces a broader definition of interoperability that goes beyond technical aspects to include legal and policy dimensions. It sets the stage for a more comprehensive discussion.

impact

This framing shaped the entire discussion by encouraging participants to consider interoperability from multiple angles, including legal, semantic, and technical layers.

We need an interoperable global approach to the sustainability of AI. So AI and energy demands are set to grow with increasing multimodal inference with the use of IoT data and with agentic AI. So we need interoperable ways to measure, to track and to incentivise better energy and water use of data centres of chips, algorithmic efficiency and data sobriety.

speaker

Sam Daws

reason

This comment introduces the important dimension of sustainability in AI governance, which had not been mentioned before.

impact

It broadened the scope of the discussion to include environmental concerns and sparked further comments on the need for a holistic approach to AI governance.

For humanity to flourish it’s vital that our diverse cultures feed into AI so we can better use it to live good and meaningful lives and that includes insights from low resource languages and also the wisdom of indigenous as people who have a minimal digital footprint, not captured by large language models trained on the internet.

speaker

Sam Daws

reason

This comment highlights the importance of cultural diversity and inclusion in AI development, bringing attention to often overlooked perspectives.

impact

It led to further discussion on the need for inclusive AI governance and the challenges of bridging the digital divide.

Trust is the most important issue. There is, AI is built on trust. And it’s not limited to the geopolitical reasons. We shouldn’t have different ecosystem. So it’s all this ecosystem are built on trust. So how to build trust? I think this is something we need to discuss.

speaker

Xiao Zhang

reason

This comment emphasizes the fundamental importance of trust in AI systems and governance, shifting the focus from technical aspects to human and social factors.

impact

It led to further discussion on how to build trust in AI systems and the role of different stakeholders in this process.

From a Global South perspective, I think the question is also really very practical as to if there are multiple fora, they only are able to invest that many resources and they might have to choose between different fora and that also can create competition between different fora.

speaker

Neha Mishra

reason

This comment brings attention to the practical challenges faced by Global South countries in participating in multiple AI governance forums, highlighting issues of resource constraints and potential forum shopping.

impact

It led to a more nuanced discussion about the need for streamlined and inclusive global governance mechanisms that consider the constraints of developing countries.

Overall Assessment

These key comments shaped the discussion by broadening its scope from purely technical considerations to include legal, policy, sustainability, cultural, and trust dimensions of AI governance. They also highlighted the challenges of creating truly global and inclusive governance mechanisms, particularly considering the resource constraints of developing countries. The discussion evolved from defining interoperability to exploring its practical implications across various domains and stakeholders, emphasizing the need for a holistic, inclusive, and trust-based approach to AI governance.

How can we develop interoperable approaches to measure, track and incentivize better energy and water use of AI systems?

speaker

Sam Daws

explanation

This is important to address the sustainability challenges posed by increasing AI energy demands.

How can we build cultural interoperability into AI governance frameworks?

speaker

Sam Daws

explanation

This is crucial to ensure AI systems reflect diverse cultural perspectives and wisdom, including from low-resource languages and indigenous peoples.

How can we improve the speed and agility of UN processes to keep up with the rapid pace of AI technology development?

speaker

Online audience member

explanation

This is important to ensure global governance mechanisms can effectively address emerging AI challenges in a timely manner.

How can we balance national AI sovereignty concerns with the need for global interoperability?

speaker

Online audience member

explanation

This is crucial to reconcile countries’ desires to develop their own AI capabilities with the benefits of a unified global approach.

How can blockchain technology be integrated with AI to address interoperability and trust issues?

speaker

Dino Cataldo Dell’Accio

explanation

This could provide a common layer of trust and transparency for AI systems.

In which specific forum should global implementation-focused AI governance coordination take place?

speaker

Online audience member (IamSarmanova)

explanation

This is important to avoid duplication of efforts and address potential geopolitical tensions in different forums.

How can digital economy agreements contribute to AI interoperability?

speaker

Neha Mishra

explanation

This could provide another avenue for addressing interoperability issues, especially at regional levels or between like-minded countries.