This week, AI policymakers, experts and developers descend on New Delhi for the world’s biggest ever gathering dedicated to the technology – the AI Impact Summit. Indian government officials estimate a quarter of a million people will join the conference and expo to discuss how to build and govern AI – Chatham House’s Digital Society Programme among them.
The summit has sky-high ambitions, and potentially great value, bringing together high-level decision-makers alongside ground-up tech builders and experts under one roof (or ten). Useful, important conversations will be had.
But it will likely fall short of producing any meaningful international governance agreement. It is too crowded a platform, with too many agendas.
To really make progress on AI governance, policymakers, developers, scientists and civil society actors interested in better governance should throw their weight behind developing solutions in smaller settings – with the opportunity to scale up them using major gatherings later.
Stormy seas for global governance
The AI Impact Summit takes place at a crucial moment. Institutions of global governance are under significant strain, as commitments to shared principles and the rule of international law give way to transactionalism.
All the while, a global AI struggle is intensifying. The US and China have the world’s most powerful tech and without their buy-in, any ambitions to reach a global agreement on AI governance are a non-starter. But US and Chinese labs are racing to the frontiers, intent on maintaining an advantage. That means AI capabilities are advancing rapidly, as is the magnitude of risk.
Recently, global scientists cautioned that real-world evidence for severe AI risks is on the rise – ranging from potential cyberattacks to biological weapons development.
In this context, other countries feel compelled to explore alternative partnerships, seeking to develop AI in way that promotes their safety and sovereignty – an approach underscored by calls for middle power solidarity like those made by Mark Carney at the World Economic Forum last month.
There has never been a harder time for global cooperation on AI governance. Nor has there ever been such urgency to figure it out. Recent years have seen valiant efforts from the UN, the G20, G7, the OECD and EU. In 2023, the UK convened the first AI summit at Bletchley Park.
Each subsequent international gathering has pushed the needle forward. But the limited, non-binding, principles-based efforts at governance made so far shrink in the face of accelerating technology and growing tensions. A global treaty on AI is a distant dream.
All roads lead to New Delhi
In this environment, international summitology might look like a dying art. But India is an enthusiastic host. New Delhi is primed for the arrival of tech CEOs and heads of state, joining closed-door high-level events later in the week.
Every few hundred meters, Prime Minister Narendra Modi’s face adorns bright posters, alongside catchy quotes (‘India has the power of double AI: Artificial Intelligence and Aspirational India’, one reads). The conference venue is huge, with queues for events and some delegates flagging down buggies to travel across the site.
The summit is a monument to India’s AI ambitions, drumming up business and showcasing tech from India and partner countries. India is anxious to maintain the resilience of its tech sector – long globally competitive but potentially threatened by labour displacement triggered by automation in key industries like manufacturing and business process outsourcing.
India is also keen to maintain sovereign control of AI – in line with its hope for strategic autonomy over foundational tech despite deep reliance on the US or China. The summit showcases sovereign and government-backed models: trained on Indian data, and hosted on an Indian cloud, an important proof of concept for a more sovereign solution.
In that respect, the summit can be seen as the latest diplomatic attempt to position India as an alternative to the US and China on tech cooperation – emphasizing equitable development and distribution of AI and the deployment of AI in developing countries. The summit may see announcements for improving Global South coordination on issues like safety. Emphasizing real-life use cases – in healthcare, energy and education – is a priority.
India also hosts the BRICS summit later in the year, and may host visits from the US and Chinese leaders – requiring a careful balancing act as it seeks to build its AI autonomy, broker Global South cooperation, and navigate the great powers. Many middle powers will face similar challenges and will hope to learn from India’s approach.
Splinter for success?
This year’s summit has already been criticized by technical experts and advocates for its content and design. The conference has incredibly broad aims rooted in seven AI chakras, including science and human capital. The summit’s main day convenes tech executives and state representatives. There are also concerns that the summit sidelines voices from civil society. Some will head to fringe events instead.
Since 2023, summit membership has bloated to include more countries, companies and organizations than ever before. Safety and security once defined the agenda. Now those issues vie for attention with promises to accelerate and diffuse ever-more-powerful AI. Such overcrowding will prevent this summit’s main track from tangibly advancing international AI governance.
This is unfortunate, but not world-ending. As AI governance experts have argued, an imperfect summit is better than no summit at all.
But policy solutions on AI governance puzzles will not come from a crowded conference room. They will come from smaller expert or regional gatherings designed to capture global, diverse inputs and operationalize them. Smaller conventions offer an opportunity to garner support for tested, trusted governance solutions, which can be scaled later in bigger gatherings.
For example, Chatham House has argued that scientist-led venues (like the International AI Safety Report, and hopefully the future UN Scientific Panel on AI) can generate a strong depoliticizing effect, building trust in scientific outputs across geopolitical chasms.
That kind of ‘splintering to scale’ is possible and, in this fragmented geopolitical environment, preferable.
Next week will see the reconvening of the independent International Association of Safe and Ethical AI’s annual conference. This is a smaller gathering, free of attachment to national prestige, with attendance from a diversity of global scientists (including from China), promising to platform technical research and regulatory solutions.
In the meantime, efforts are underway to build agreement over digital technical standards for AI safety. Civil society groups are convening scientists and experts to seek agreement on a set of ‘red lines’ for unacceptable AI risks, like bioterrorism. Governance labs and academics in India and globally are exchanging ideas on participatory governance solutions to long-standing tech policy problems.
The value of these smaller efforts is their potential to produce globally scalable outputs with expert buy-in – championed by consensus-building states, like Singapore, Brazil or Switzerland. They can create trust in mechanisms and process, laying the groundwork for consensus on principles for AI governance (an optimistic objective) and implementing pragmatic governance solutions (a more likely aim).
But they must also be wary of replicating power asymmetries that are so often a stumbling block to effective AI governance.
Nightmares of inaction
AI governance hopefuls face a steep hill. The question is not whether to scale trusted solutions, but how. Political appetite for global cooperation on AI is waning. The US did not back the AI Safety Report and has sought to remove guardrails on AI development, as it competes bitterly with China for influence over global AI.