Research Director: Philip Tomei

Coordination Infrastructure for Preserving Human Agency in the Age of AI

One of the most under-examined systemic risks associated with advanced AI is the gradual erosion of human agency across economic, political, and cultural systems. Even in the absence of catastrophic AI failures, competitive pressure to deploy increasingly capable AI systems can progressively weaken both explicit mechanisms of human control, such as voting, consumer choice, and executive authority, and the implicit alignment that institutions maintain through their reliance on human participation.

This dynamic, which we describe as gradual disempowerment, is introduced in our ICML 2025 paper and continues to generate international discussion. Over time, these shifts may produce a form of civilizational drift in which key social systems operate increasingly independently of meaningful human direction. Once certain thresholds are crossed, these dynamics may become difficult or impossible to reverse. At present, however, there are no systematic measurement frameworks, monitoring systems, or coordinated research efforts focused on detecting and mitigating these risks.

The Gradual Disempowerment Observatory addresses this gap by providing shared infrastructure for research, measurement, and coordination. The Observatory functions both as a data platform and as a collaborative hub for researchers, policymakers, and institutions working to understand how AI deployment reshapes human influence within major societal systems.

At its core, the Observatory maintains a continuously updated dashboard that tracks indicators related to human agency across interconnected domains. More broadly, the project expands and coordinates research on gradual disempowerment by lowering barriers to data access and fostering collaboration across disciplines.

The Observatory includes:

• Compound indices for measuring changes in human agency across economic, political, and cultural systems

• Automated data pipelines that aggregate and update relevant indicators from diverse public and institutional sources

• Threshold detection models that identify potential tipping points where feedback loops accelerate disempowerment dynamics

• Cross-domain analysis tools that examine how developments in one system, such as economic displacement, influence political power or cultural influence

• Open data standards and APIs that allow external researchers, policymakers, and institutions to contribute metrics, validate findings, and conduct comparative analysis across regions and sectors

Together, these tools create a shared empirical foundation for understanding and responding to the societal impacts of advanced AI.

Beyond the data infrastructure itself, the Observatory supports broader coordination across the emerging research field. This includes:

• Developing an evidence base for governance, including regular reports analyzing how different patterns of AI adoption affect human agency and institutional responsiveness

• Supporting interdisciplinary research collaboration by providing a shared platform for datasets, methodologies, and findings

• Convening researchers and policymakers through workshops and roundtables with existing collaborators, including engagements associated with Chatham House, NeurIPS, and ICML

• Supporting early-career researchers through fellowships and PhD scholarships focused on gradual disempowerment and related governance challenges

• Improving public communication by translating abstract long-term risks into measurable societal trends through the Observatory’s data visualizations and dashboards

By providing shared measurement tools, open data infrastructure, and a coordination platform for researchers and policymakers, the Gradual Disempowerment Observatory strengthens society’s capacity to monitor and respond to long-term shifts in human agency as AI systems become increasingly integrated into economic and institutional life.