Global Dispatches
Global Dispatches
A Summit in Barbados Offers an Opportunity for Global Cooperation on AI Safety
0:00
-25:15

A Summit in Barbados Offers an Opportunity for Global Cooperation on AI Safety

As the world grapples with the risks of AI, a summit in Barbados aims to turn international dialogue into practical cooperation.

Next week, government officials, technology leaders, and civil society representatives will gather in Bridgetown, Barbados, for a special summit on AI safety. The meeting comes amid growing questions about how international cooperation can help mitigate the global risks posed by a technology increasingly dominated by just a handful of countries and companies.

The United Nations is co-hosting the summit alongside the government of Barbados, but the goal is not to negotiate formal rules or establish a new international regime for AI governance. Rather, the aim is to bring governments, technology experts, and other stakeholders from around the world into the same room to identify practical ways they can work together to address known and emerging risks from AI.

I’m joined by Amandeep Singh Gill, the United Nations Under-Secretary-General and Special Envoy for Digital and Emerging Technologies, who explains why this meeting is taking place, why Barbados is hosting it, and what organizers hope to accomplish. We also discuss the broader challenge of fostering international cooperation on AI safety at a time when the technology itself is advancing rapidly.

The episode is freely available above and wherever you get your podcasts, including Spotify and Apple Podcasts.

Share

Transcript edited for clarity

Mark Leon Goldberg: Welcome to Global Dispatches, the podcast that helps you understand the world through conversations with the people who know it best. I’m your host, Mark Leon Goldberg, an international affairs journalist who’s been covering global politics for more than two decades. Enjoy the show, and please visit globaldispatches.org to learn more, sign up for our newsletter, and support independent journalism with a global perspective.

I am joined today by Amandeep Singh Gill, the United Nations Under-Secretary-General and Special Envoy for Digital and Emerging Technologies. Welcome, Mr. Under-Secretary-General.

Amandeep Singh Gil: Thank you, Mark, for having me. Please call me Amandeep.

Mark Leon Goldberg: Okay, well, Amandeep, we are speaking ahead of the Bridgetown Special Summit on AI Safety taking place in Bridgetown, Barbados. To kick off, what prompted the UN and Barbados to convene this special summit on AI safety? What’s the backstory here?

Amandeep Singh Gil: The backstory is the concern that’s been rising with some of these incidents where AI models have hacked into websites, including public websites, and have displayed intent, taken actions which are not aligned with the goals that human beings have given them. So, at the General Assembly in September, 135 countries mentioned AI, including its risks. This is like up from close to 50 such mentions in 2025, and only three such mentions in 2017.

Mark Leon Goldberg: And you’re referring to the speeches that leaders give from the General Assembly podium, that the number of AI mentions has surged. So there’s interest here.

Amandeep Singh Gil: There is interest. There is concern. And I think what we seek to do with this special summit on AI safety that the UN Secretary General Antonio Guterres and Prime Minister Mia Mottley would co-host is to focus attention on where international cooperation can make national approaches stronger, more compatible with each other, and more effective without imposing a single model. And, of course, national action is indispensable, but international cooperation is needed because AI risks and impacts cross borders, and the capacity of different countries to assess and manage them remains deeply uneven.

And I think there is another point that I want to underline here why this conversation is important. Often these conversations are focused on risks coming from the so-called frontier models, coming out of a few labs in the United States or in China. Those risks are important, but these risks, overall, do not arise only at the frontier. So wherever AI is deployed without adequate safeguards, accountability, or understanding of local context, these risks can arise. So we want to focus attention through this special summit on this broader view of AI risks.

Mark Leon Goldberg: And why Barbados in particular? I mean, Mia Mottley, the Prime Minister of Barbados, is known as a champion of multilateralism, that’s for sure. She’s gained a lot of prominence on climate change, debates in particular. But why Barbados for a summit on AI safety? So

Amandeep Singh Gil: Barbados has consistently argued that global challenges, you mentioned a couple of them, and this AI risk issue and the AI divide is a global challenge. So, Barbados says these global challenges require solutions that work for everyone. And countries that are most exposed to the global impact of these issues should not simply adapt to aqueous decisions made elsewhere.

They should help shape them. So I would say this is a great venue holding this summit in Barbados, helps us to broaden an AI safety debate, which is often centered only on countries developing the most advanced systems.

Mark Leon Goldberg: Walk me through this meeting. What can we expect?

Amandeep Singh Gil: The setting is a focused discussion among governments, technology companies, scientists, civil society, and international organizations. So, there will be an opening segment, which the Prime Minister and the Secretary General would address. We are expecting messages of solidarity and sharing of perspectives by leaders from different parts of the world. I believe from every continent, and also expecting the participation in person of senior level ministers from some of the countries that have been recently leading on this issue, especially Finland and Norway that issued a call for the control of frontier AI during the High-Level Week.

And then the participants will cover these issues in detail through two roundtables, One focused on multilateral AI risk management, including evaluations of these models, learning from incidents, preparedness, and national capacity. And the second one will look at issues of excess, inclusion, diffusion, linguistic and cultural representation. In other words, how to make AI work for all of humanity.

Mark Leon Goldberg: And then in terms of outcomes, like what to you, being so close to this issue, would represent a good or useful outcome from this particular meeting?

Amandeep Singh Gil: We are, in fact, looking for greater clarity on what could be practical elements of a multilateral approach, a multilateral framework on AI risk management, which is not just nice words, but which is something that’s backed by independent and credible oversight. The Secretary General made a specific reference to this type of an approach in his remarks to the UN Journal Assembly in September.

So, the choice of this kind of an approach means that this is not uniform global regulation. Different countries will make different choices, but we could have some common features, including the shared understanding of emerging risks, comparable evaluations, common terminology, some learning from serious incidents, so sharing of information, transparency around serious incidents, near misses, stronger national evaluation capacity, so what kind of national evaluation capacity is needed.

So, giving some indication of that, and in a sense, making the case for greater resource allocation to building this capacity, and safeguards then that are interoperable, independently verifiable, and rooted in our shared norms at the United Nations.

So in a sense, if I can just put it in one sentence, it’s like something that allows the world to come together regardless of whether you are a frontier country hosting frontier model development or country applying open source or other models that are developed elsewhere. You have a shared understanding of how these risks arise, what kind of risks you should prioritize, what kind of measures you should take to address those risks. And if there are residual risks, you know, do you manage them through insurance schemes?

Do you strengthen national resilience with the assumption that some risks would materialize and you would have to do something? So it gives us a common framework, common sheet across different contexts to reinforce AI safety and trust so that diffusion can take place and the application of AI, the benefits of AI can be realized.

And so if I am understanding you correctly, this meeting in Bridgetown is essentially an opportunity to start talking about those practical elements of multilateral cooperation that you described earlier. Not some sort of overarching global governance treaty on these things, but what are the discrete opportunities that countries, governments, private sectors, civil society work together on AI safety?

And you listed a number of those. I suppose then, being like a kind of UN nerd myself, what’s the process for getting there? Because at the UN, process often dictates outcomes. So how are you seeing this particular meeting in Barbados leading to the actual implementation or the definition and implementation of some of those practical elements of cooperation that you articulated? So Barbados is the starting point. It’s one step in a continuing process.

I mentioned the expression of concerns and the unveiling of certain calls and proposals by member states. Now, Bridgetown, Barbados, would help maintain the momentum and clarify some of these issues, build some common understanding. And next year, we have very important opportunities coming up to land these common understandings into some form of practical action. So, at the far end of the year, in December, you have the G20 Summit, which is hosted by the UK in Manchester.

You have a summit that’s hosted by Switzerland on AI in June. And importantly, in May, you have the second global dialogue on AI governance at UN headquarters in New York. So there are these multiple opportunities for governments and other stakeholders to come together and act on these ideas. So Bridgetown is an important way station, but it’s near not the end point of this process.

Mark Leon Goldberg: So there are other opportunities to build on whatever momentum is gained, whatever agreements, informal or otherwise, might be made in Barbados, that the next time people meet to discuss these issues, they’ll be able to progress upon those previously agreed-upon ideas.

Amandeep Singh Gil: 100%. So if we identify areas of convergence, they can feed into practical work on evaluation of these risks, interoperability in terms of safety infrastructure, learning from each other, information sharing, national preparedness, national capacity. And these areas of convergence and the possibilities for action on those areas can be taken forward in the light of existing UN mechanisms.

So you have the Independent International Scientific Panel on AI, which, by the way, is meeting just after this particular summit. So it has published a thematic brief on AI safety, looking at some of these recent incidents. It’ll continue its work providing independent scientific advice to policymakers on risks and opportunities and impacts. The dialogue allows policymakers from all parts of the world to come together and find common ground. And these summits, whether they are this AI summit or the G20 meetings, can allow coalitions of countries to take action, take decisions.

So this continuity, this interplay between independent science, inclusive policy discussions, and practical measures on AI risk management is what we are looking for. And in a sense, you are a UN nerd. You know that these days, centralized, top-down approaches are very difficult to pursue at the UN. So this kind of a networked multilateralism, where you have a kind of an inside-outside dynamic in terms of internal UN processes and coalitions acting outside, including multi-stakeholder coalitions, could perhaps allow us to deal with this rapidly evolving issue in an agile and effective manner.

Mark Leon Goldberg: So I’m glad you brought up the independent scientific panel on artificial intelligence. This is a panel that’s less than a year old. It’s co-chaired by Maria Ressa, the Nobel Peace Prize-winning journalist from the Philippines, and Yoshua Bengio, the Canadian computer scientist, one of the kind of godfathers of artificial intelligence and someone who’s been very much on the floor of discussions on AI safety issues. What’s your assessment of the impact that this panel has had in the less than a year that it’s been in existence?

Amandeep Singh Gil: The impact has been significant. I think their preliminary report that was released in July at the first global dialogue on AI made those policy discussions concrete and evidence-based. And countries cutting across different regional groups, you know, often they don’t agree with each other on what is the best way to pursue AI governance. All of them, in fact, all these regional groups agreed on the importance of this work and encouraged the panel to go further. And they are now beginning work on their full annual report, which would inform the next dialogue.

Of course, this is an independent panel. We have 40 independent experts from all five UN regional groups. They serve in their personal capacity. Their job is to provide a shared scientific evidence base of AI’s capabilities, opportunities, risks, and impacts, identify areas of scientific agreement, uncertainty, even gaps in evidence. And it’s important to note that the panel does not negotiate policy, does not represent governments or companies or make recommendations.

And the meeting in Bridgetown on 23rd November will not direct its work or conclusion. But for the UN to have a standing capacity, independent capacity, which we have in other areas like IPCC on climate change or the International Resources Panel serviced by the UN Environment Programme, I think it was important.

And the distinction that this panel has from other such institutions in the UN context, in terms of these outcomes are not negotiated, politically negotiated outcome, they really represent independent scientific advice -- this is now increasingly appreciated. The panel has moved into a more active phase of its work. It issues this thematic brief on AI safety, and it is working on other thematic briefs, issues like AI and the environment, AI’s past building requirements, and these would help governments around the world follow the evidence, follow the science, and then take the right decisions.

Mark Leon Goldberg: So I wanted to go back to something you mentioned earlier about different perceptions of what constitutes AI risk. And your position, your role, I think puts you in a unique place insofar as you are probably one of the very few international officials who has conversations about AI risks with virtually every constituency on the planet. You’re speaking to governments and civil society people in the developing world. You’re speaking to folks in Silicon Valley. How are your conversations about AI risk different in different contexts in which you’re meeting people? Or are they generally the same?

Amandeep Singh Gil: No, I see a broad agreement on the need for safe, reliable, and accountable AI. But different countries around the world are experiencing this technology from different perspectives. So, they are at different starting points. And so often in developing countries, I see more of an emphasis on exclusion, unequal access, reliance on imported systems, the impact on employment, linguistic and cultural suitability, and their capacity to incorporate, evaluate technologies that are developed elsewhere.

Mark Leon Goldberg: So, their idea of risk is they’re getting risked being left behind in this kind of global technological revolution.

Amandeep Singh Gil: 100%. So often I hear like the first industrial revolution left us by the wayside, in fact, led to colonialism. And then you have the second one, third one -- we don’t want to miss this one. We also need to be in on AI.

Mark Leon Goldberg: That’s interesting. I should say I heard that from William Ruto at UNGA this year, the president of Kenya, who made a very similar argument to what you’re making, that we can’t let this be a new form of colonialism that extracts our data, our information, and we’re left behind.

Amandeep Singh Gil: Yes, so that’s a concern I hear. I also hear concerns about, and this is often the case in Europe, in North America, that this will lead to a huge concentration of economic power in the hands of a few companies, a few individuals. It will worsen the situation within society in terms of equity.

And there is a kind of a diffuse anxiety also about the future in terms of employment prospects. What happens to my job? What kind of education, learning should I be pursuing to be fit for purpose in the AI age? So there’s this diffuse anxiety, and it’s becoming a political issue in many ways. We should also not ignore that some of the leading developers themselves are talking about more systemic risks, even existential risks.

Systemic risks in the sense of how, if we rely more and more on these systems in our financial markets, for example, in national security, in warfare, what could go wrong? So they’re looking at scenarios that could go wrong and what kind of impact that would have on our societies and on economies. And some people even say that this technology could wipe out humanity. I don’t believe it myself.

So there’s no evidence; there are no pathways today that you could conceivably see leading you to that kind of an outcome. But I think we must acknowledge all these different types of risks and they are spread out geographically in terms of their differentiation, but they’re also spread out over time. For us at the UN, what really matters is the short-term and the medium-term impacts, the first-order impacts in a sense, and the second-order impacts in terms of inclusion, exclusion, inequities getting reinforced on top of existing inequalities, labor market impact, institutions, public sector institutions not being able to adjust.

And of course, in the short-term, we have to sort out the cybersecurity vulnerabilities misused by non-state actors to develop bioweapons, for example. So those are kind of like some of the issues that we see.

Mark Leon Goldberg: So a year from now, and this could be my last question, say we’re speaking again. What do you have hoped would have happened through a variety of these multilateral processes that you just discussed, starting with the meeting in Barbados? And then also you mentioned a meeting in Switzerland, the Global Dialogue at the United Nations, and then the G7. Like a year from now, where do you expect or hope the state of multilateral cooperation on these issues to be?

Amandeep Singh Gil: It would be nice within the next one year, 12 months, to come to a common understanding on what kind of risks we should be prioritizing, how should we be assessing those risks, and each country would need to assess risks in their own context. It would be good to have, alongside, a common understanding on what kind of a national response is needed from these evaluations to audits to some of the resilience measures that need to be taken.

Our cyber systems are quite vulnerable to these agents operating at scale, doing good things, but then they can misbehave, and they can get out of hand. So, having that common understanding. And most importantly, having these channels for international cooperation, where good safety practices flow across borders, where we have common understandings of how evaluations are done, where some of these independent evaluators are recognized across borders.

So there is a mutual recognition of audits and evaluations, where we have transparency in terms of reporting on incidents, sharing of information so that your counterparts in other countries can take measures to protect themselves. And where, in a sense, we have some early warning mechanism where the risk is systemic, where the risk is going to be substantial beyond the capacity of a single country, ountries are in a position to come together and cooperate with each other.

And this is not new because it has happened in many other domains before. Take the nuclear domain, for example, nuclear safety and security. We have a common understanding of the severity of nuclear safety incidents. And then we have an understanding of how nuclear plants should be built, defense in depth, what kind of containment measures.

We have an understanding of how people should be trained to operate them. We also have some common understanding on how to manage cross-border liability on these issues. What is the risk that you can contain? What’s the residual risk that needs to be managed through insurance arrangements, liability arrangements? And are there kind of some international pooling mechanisms, international response mechanisms? So, I think AI is going to be significant force in our societies and in our economies. So we need to prepare just as we prepared in the past for civil aviation, in the financial sector, in the nuclear sector.

So you can see I’m going in the direction of my thinking is this is an economically significant force. It is going to pervade societies. These models are going to be used in many, many areas. So why not prepare ourselves institutionally at the national level, at the international level, so that risks are contained and we can focus our minds, our resources more on the opportunities, more on the dividend that these technologies offer for the sustainable development goals.

Mark Leon Goldberg: Amandeep, thank you so much for your time.

Amandeep Singh Gil: Thank you, Mark. Thank you for having me.

Share

Discussion about this episode

User's avatar

Ready for more?