The co-facilitators from Spain and Costa Rica have published the Zero Draft for the modalities of the Independent International Scientific Panel on AI and the Global Dialogue on AI Governance. In our analysis this is a good starting point. At the same time, we also want to highlight three suggestions: 1) The focus of the working groups of the panel would benefit from more clarity, 2) the Advisory Committee should fully leverage the UN’s universal nature, 3) the Advisory Committee’s role should be in review, not in drafting the scientific synthesis.
A good, constructive start
We appreciate the work of the co-facilitators. Bridging all the different views is a complex and challenging task. The zero draft is a constructive step towards a panel and a dialogue that deliver. As we have highlighted in our recommendations it is crucial to have a panel with capacity, political legitimacy and scientific independence. Our analysis of various institutional models also shows that a multi-tiered structure is the most efficient way of achieving this. The zero draft aims to achieve legitimacy, independence, and capacity with a combination of an Advisory Committee elected by the General Assembly, an independent Expert Committee steering the scientific work, and dedicated working groups. We encourage Member States to preserve this structure and refine its functioning throughout the negotiations.
Similarly, for the dialogue, we appreciate the effort of the co-facilitators to balance demands for an intergovernmental exchange with those for a pure multistakeholder format, and to choose co-located events in a way that engages the two main UN hubs in Geneva and New York. As highlighted in our own analysis, AI for Good in Geneva has the strongest existing engagement in the ecosystem with all relevant stakeholders, but it is not as developed with regards to high-level government-to-government exchanges, where the High-Level Week in New York is particularly strong. We also strongly support the choice of initiating the first dialogue in September 2025, giving an opportunity to have a global conversation on this year’s rapid AI developments.
The need for more details on panel topics
A UN modalities resolution has no fixed required level of detail. Indeed, there is precedent for countries only defining high-level outlines and leaving it to the implementing bodies (Advisory Committee, Expert Committee, Secretariat) to define more granular terms of references. Having said that, we do encourage member states to suggest more details with regards to the specific topics that they want the panel to address.
The zero draft restates the mandate of the panel to cover opportunities, risks, capabilities, and impacts of AI. However, given that AI is a general-purpose technology that intersects with just about every other topic, the specified scope is still too broad. We encourage countries to be more granular and to highlight the specific topics that they would like to see covered in working groups (see second figure in our recommendations. Where applicable, we also encourage countries to highlight specific institutions external to the UN that have relevant expertise on suggested topics, which could support such a working group.
- Nominating and selecting for the relevant expertise: Scientific expertise is specialized. Having a clearer sense of focus areas helps inform an adequate nomination and selection of experts for the Expert Committee.
- Policy relevance: As multiple stakeholders have highlighted; this report should not be a purely academic exercise but should be relevant for diplomats and policymakers. Getting a sense of what questions or topics are in demand is important to ensure policy relevance.
- Depth vs. breadth vs. speed: An unfocused panel that tries to cover everything but does not reach the required depth, would not be as useful as working groups that produce state-of-the-art summaries on specific areas. Similarly, requiring both depth and breadth can lead to ballooning complexity and longer report production timelines as evidenced by the historical slowdown of IPCC report cycles.
Aside from more details on the focus of working groups, we also suggest that in the sentence “invite external experts to engage with the working groups as necessary” the words “engage with” are replaced by the words “be part of” to reduce ambiguity.
Universal membership is more legitimate and not necessarily slower
The zero draft suggests an Advisory Committee limited to forty members elected by the General Assembly to equally represent the five UN regions. Keeping the Advisory Committee at a manageably small size might make intuitive sense. However, it makes the process more politically fraught and plausibly even slower. Below we highlight why we strongly recommend a universal membership model.
We think that manageable group size is crucial for bodies that draft language, such as it is the case for safety standards in ICAO or the IAEA. In contrast, for bodies that provide feedback and review on a specific draft such as the IPCC and IPBES plenaries, the exact group size is less important. We are familiar with concerns from stakeholders that a universal Advisory Committee could water down scientific independence and/or slow down the panel. However, based on both research and expert interviews, we find that:
- Political pressure to amend scientific findings has historically primarily come from large countries. Experience from other scientific panels indicates that it is primarily countries with significant resources and interests that have systematically “lobbied” to amend scientific findings based on a political logic. Anecdotally, we note that it was a governmental expert from a country with limited resources and limited instructions from capital who has been crucial in protecting what is arguably the most important scientific finding in IPCC history (see Stephen Schneider. (2009). Science as Contact Sport. pp. 138-141)
- Political approval is not the main bottleneck to produce a global assessment report. We have conducted interviews specifically geared toward understanding the bottlenecks for efficient scientific synthesis, and we found that consensus on outlines, unbounded scope, repeated calls for nominations to match demographic desiderata, and multiple rounds of review are all more significant time factors than final approval. As an example, in the IPCC production timeline final approval is a low percentage of the overall time, and this is despite a very demanding approval process that includes consensus by plenary for both working group reports and summaries for policymakers (!).
- There are ad-hoc ways to create manageable group sizes for the few cases where negotiation might still be needed. For example, if the plenary review hits an impasse in its approval process, contact groups with the lead authors of that section and core countries can get together to discuss the science behind it, and the exact wording.
Recalling that the final report of the High-level Advisory Body on AI has framed the scientific panel as an inclusive process that fills a specific gap by engaging all UN member states. We still think that universal membership in the Advisory Committee is desirable and preferred.
The need for clear functions for the Advisory Committee
The current wording on the role of the Advisory Committee is that it “shall contribute to the Panel´s outputs led by the Expert Committee”. This is vague and opens the door to too many interpretations that would undermine the scientific independence or political relevance of the panel. In line with our recommendations, we suggest the following:
- The Advisory Committee should provide input and review: We see the role of the Advisory Committee in providing political legitimacy. On the one hand, this includes providing non-binding inputs on what topics the panel should cover, and nominating relevant experts. This is an important demand signal to ensure that the work of the panel is policy relevant and that there are no blindspots. On the other hand, this includes the review of the main annual report produced by the panel. This review should include a final approval of the summary for policymakers and the ability to note dissent on specific sections. This ensures that scientific witnesses from many regions can attest the transparency and validity of the report’s findings. Without review the engagement and trust of member states with the panel’s outputs will be lower.
- No direct involvement of the Advisory Committee in working groups: We suggest the sentence “assign members of the Expert and Advisory Committees to each working group” for deletion. Including experts elected by the UN General Assembly to equally represent geographic regions in the working groups mixes functions and directly undermines the scientific independence of the panel.
- The Advisory Committee should consist of governmental experts: We note that the ideas that the Advisory Committee should both be elected politically and represent all UN groups equally but also be experts serving in personal capacity are mutually incompatible. We suggest that the actual drafting of the synthesis reports should be conducted by independent experts selected based on scientific merit and serving in personal capacity. In contrast, the Advisory Committee should consist of governmental AI experts.
Overall, this is a good start. Still, to ensure that the scientific panel efficiently delivers the full value to member states, the devil can be in the details. We encourage all stakeholders to continue the constructive engagement that has characterized this process. Please do not hesitate to reach out to kevin@simoninstitute.ch with comments or questions.