In September 2024 the UN General Assembly adopted the Global Digital Compact, mandating the establishment of an Independent International Scientific Panel on AI and a Global Dialogue on AI Governance. The next step in this process will be to negotiate the modalities on how exactly to set-up these institutions.
Here at the Simon Institute (SI), we have closely followed the negotiations leading up to the Global Digital Compact and are committed to providing transparent and independent research to ensure its effective implementation. In this context, we are sharing a first interim report on a) trade-offs in institutional design that Member States will have to weigh, b) the building blocks of these novel institutions, and c) potential archetypes for arranging these key building blocks.
The goal of this first interim report is not to prescribe any specific institutional model but to enhance understanding among Member States and other stakeholders about design options.
Trade-offs
Scientific assessment vs. policy relevance: The panel will likely aim to produce scientific assessments that are policy-relevant without being policy-prescriptive. Scientists often prioritize rigor, whereas policymakers require timely contextualized recommendations for action (e.g., the need for rapid scientific synthesis during the COVID-19 pandemic). Potential solutions include standardized methods for conveying levels of uncertainty, like in the IPCC, as well as channels allowing policymakers to request scientific analyses.
Technical expertise vs. geographic representation: While involving the world’s best scientists (often from developed countries) ensures expertise, a commitment to “balanced geographic representation” is required to build trust among nations. Potential solutions include travel grants and other forms of participation support for scientists from developing countries. Beyond that, there is potential for broader capacity building. This can also include collaboration among developing nations to pool capacities, such as establishing regional AI safety institutes, rather than spreading resources and talent more thinly.
Inclusion vs. decision-making capacity: Broad stakeholder inclusion can enhance legitimacy and technical expertise but can hinder efficient decision-making. A full multi-stakeholder body akin to the Internet Governance Forum can facilitate dialogue but may produce fewer tangible outcomes due to limited decision-making abilities. Potential solutions include intermediate forms of multi-stakeholder participation. For example, a panel could allow civil society and private sector participants as observers without voting rights or adopt weighted voting models like the International Labour Organization’s tripartite system. Further, conflict-of-interest policies may be useful to monitor private sector involvement and screen scientific authors.
Global assessment vs. local knowledge and values: While AI’s cross-cutting impacts necessitate a range of scientific disciplines, disciplinary over-inclusiveness can dilute focus. The IPCC’s success partly stems from concentrating on domains with global scope and measurability, despite criticisms of limited social science integration. A potential solution is to balance the focus of the panel with the focus of the dialogue: prioritize areas of global relevance, such as macroeconomic impacts and global risks for the panel; and address topics with more ambiguity through an inclusive dialogue, leveraging civil society’s key strengths.
Building blocks
Establishing the Independent International Scientific Panel on AI and the Global Dialogue on AI Governance involves critical decisions across several key building blocks:
- Membership: Determine participation criteria for governments and stakeholder groups, including the roles and rights of observers and non-state actors.
- Governance structure: Define mechanisms for political and scientific steering, including decision-making processes and the roles of steering committees and technical working groups.
- Funding: Decide on funding sources—whether through assessed contributions, voluntary donations, or multi-stakeholder funding. Consider capacity building to ensure balanced participation and whether or not experts are remunerated.
- Panel-specific options: Establish the panel’s thematic focus, scope of research (synthesis vs. original research), potential working groups (e.g., frontier AI capabilities projections, macroeconomics of transformative AI, CBRN amplification risks), and processes for report writing, author selection, and review.
- Dialogue-specific options: Design the format and agenda for the Global Dialogue, including the number of tracks, stakeholder engagement levels, and integration with existing UN conferences and meetings.
Archetypes
To navigate the design complexities, we have defined a few institutional archetypes for both the Scientific Panel and the Global Dialogue:
For the panel:
- Representative, governmental expert panel: Each UN member state nominates experts, ensuring broad geographic representation but potentially compromising scientific expertise and decision-making efficiency.
- Representative, multi-stakeholder panel: Involves representatives from governments, private sector, academia, and civil society, promoting inclusivity but possibly diluting scientific rigor and complicating consensus-building.
- Independent scientific panel with political approval: Experts are selected based on merit, with political approval of reports to ensure policy relevance and legitimacy, balancing scientific independence with governmental endorsement.
- Independent scientific panel with political relevance: Maintains scientific independence without formal political approval, focusing on engaging policymakers to ensure relevance but potentially facing challenges in political buy-in.
- Independent scientific panel within an NGO framework: Operates outside governmental structures, maximizing scientific independence and agility but possibly lacking political legitimacy and influence on international policy.
For the dialogue:
- Multi-stakeholder dialogue: An inclusive platform engaging all relevant stakeholders, fostering diverse perspectives and knowledge sharing but potentially lacking decision-making power and efficiency.
- Pre-standardization dialogue: Focused on technical experts developing global norms and standards, promoting innovation and technical solutions but possibly excluding non-technical stakeholders and broader societal considerations.
- High-level political dialogue: Involves only government representatives, facilitating policy negotiations and high-level engagement but excluding other stakeholders and potentially limiting transparency.
- Multi-track dialogue: Combines elements of the above models with multiple tracks for different stakeholder groups, promoting comprehensive engagement and cross-pollination of ideas but requiring significant coordination and resource investment.
You can read the full interim report here. Additional research on analogies and the international AI governance can be found in the separate Annex here. For questions and comments, please reach out to Kevin Kohler at kevin@simoninstitute.ch.