Future Brain Trust

First, a shared language for describing AI systems, especially at the level of structure rather than just behavior. Today, we lack stable abstractions for comparing models across institutions and jurisdictions. Even a preliminary step toward common representational frameworks, or ways of describing how systems organize meaning internally, would be a major advance for both interpretability and governance. Second, it should acknowledge AI as part of critical global infrastructure, not just a technological sector. This reframing shifts the conversation toward resilience, dependency mapping, and systemic risk, aligning AI governance with how we already think about energy, finance, and/or supply chains. Third, the creation of interfaces, between models, institutions, and worldviews.

• Interoperability of governance approaches
• Open-source software, open data and open AI models
• Transparency, accountability, and human oversight

AI systems are inherently transnational, while governance remains fragmented across jurisdictions. Without interoperability of shared standards, audit mechanisms, and regulatory interfaces, we risk creating isolated "governance silos" which cannot effectively manage cross-border systems. Current AI systems operate with limited interpretability, creating gaps in accountability. Advancing transparency is not only about explainability at the output level, but about developing frameworks to understand how systems structure decisions internally. This is critical for enabling meaningful human oversight, especially in high-stakes domains. AI governance must be anchored in human rights as a non-negotiable baseline. However, this requires translating abstract rights into operational constraints, how rights are encoded into system design, evaluation, and deployment. Open ecosystems are meant to both democratize access and enable independent scrutiny. Open models and datasets are essential for building public-sector capacity, fostering innovation, and supporting auditability.

Behavioral adaptation and feedback loops between AI systems and human actors are insufficiently captured. AI does not operate in a static environment; it co-evolves with user behavior, institutional incentives, and regulatory signals. This creates second-order effects such as strategic gaming, norm shifting, emergent coordination failures, and even models that try to manipulate; these require governance models capable of learning and adapting over time. Environmental externalities of AI systems, including energy use, water consumption, and material supply chains. These should be integrated more directly into governance discussions, especially given their global and long-term implications.

A primary challenge is the misalignment between technical capability and institutional readiness. While AI systems can generate high-resolution environmental insights (e.g., nutrient tracking, water quality forecasting), many public agencies lack the capacity to validate, integrate, and act on these outputs. This creates a paradox where decision-makers are either over-reliant on opaque systems or underutilize valuable intelligence altogether. Second, fragmented governance approaches across jurisdictions hinder interoperability. Environmental systems are inherently transboundary, yet data standards, model evaluation protocols, and regulatory expectations remain inconsistent. This limits cross-border collaboration and weakens collective responses to shared risks such as water scarcity and ecosystem degradation. These gaps also present significant opportunities. There is growing momentum to develop shared evaluation frameworks, open benchmarks, and federated data infrastructures, which could enable more coordinated and transparent governance. Advances in privacy-preserving computation and distributed learning also create pathways for collaboration without requiring full data centralization.

Translate fragmented national approaches into interoperable governance systems by establishing shared standards, enabling cross-border accountability mechanisms, and fostering sustained institutional collaboration around AI as critical global infrastructure.

It can catalyze joint pilot projects, such as 1Treelion.org, from shared audits, cross-border sandboxes, to evaluation infrastructures, that move governance from static frameworks to iterative, evidence-based practice. The Dialogue's role is to transform a landscape of parallel efforts into a functioning global governance system.

Panels and working groups composed of institutional, university, and private labs focus on specific domains. Also delegates from every region's largest private AI companies should be organized to share with public their work.

Independent technical researchers and auditors, especially outside major labs, lack access to models, datasets, and evaluation infrastructure. This limits meaningful external scrutiny. Future-oriented and interdisciplinary voices (e.g., systems thinkers, complexity scientists, long-term risk researchers) are often marginalized in favor of short-term regulatory concerns. These perspectives are essential for anticipating systemic risks.

I would love to see some sort of on-going "panel with the public" happening as a "side-event", where delegates sit and dialogue with live questions from social media. This would be broadcast on a social media platform.

n/a