As governments experiment with AI to simulate public opinion, new questions are emerging about who these systems truly represent. This Research Radar examines the Collective Intelligence Project’s Digital Twin Evaluation Framework—a developing method for testing whether AI models can mirror real human opinion patterns—and explores the democratic risks of relying on synthetic publics in policymaking.

This week, our new Research Radar warns that the White House’s Genesis Mission to merge federal AI, data, and supercomputing power could sideline universities, communities, and public oversight. In a

As governments and international bodies race to establish guardrails for AI, most of the global agenda still focuses on managing what AI systems produce—their outputs. This article argues that such an approach is incomplete. The real foundations of safe, rights-respecting, and equitable AI lie upstream in how data is collected, governed, shared, and stewarded. Without integrating mature data governance practices, such as data stewardship and data commons, into AI governance, countries will struggle to protect fundamental rights or ensure that AI’s economic and social benefits are distributed fairly. A future-ready AI governance framework must therefore unite input and output governance into a single, coherent system.

The Trump Administration’s Genesis Mission aims to unify federal supercomputers, datasets, and AI systems into a single national platform for scientific discovery. But as Beth Simone Noveck argues, the plan centralizes unprecedented research power while offering almost no role for universities, communities, or the public, raising urgent questions about access, transparency, and democratic accountability.

AI is already shaping government decisions, but bright-line bans on automated decisions are not enough to manage the complex “middle space” where most automation now operates. Effective governance requires both individual rights — so people can understand and challenge decisions about them — and collective oversight that lets journalists, civil society, and regulators scrutinize systems as a whole. By blending transparency, redress, and sustained human responsibility, governments can harness AI to improve services while safeguarding democratic accountability.

Political Watch’s “What Our Representatives Do” tool (QHLD) organizes thousands of congressional initiatives into clear, searchable themes, making political priorities visible, comparable, and actionable. By exposing imbalances like the 198 initiatives on squatting versus just 54 on social housing, QHLD shows how structured, understandable data can reconnect citizens with their representatives. Yet the project also highlights a global challenge for civic tech: limited funding, slow adoption, and the high cost of responsible AI integration. Making democracy legible is possible, but only if we invest in tools designed to share power, not just document it.

The Solving Public Problems course has helped learners worldwide tackle complex challenges. The course teaches how to leverage technology, data, and collective wisdom in our communities to design powerful solutions to contemporary problems. Now, Beth Simone Noveck is exploring how to remake it for the AI era, using technology to make problem-solving skills easier to learn without losing the human connection at its core. Your input is needed!

Princeton’s Mihir Kshirsagar breaks down why predictive-policing algorithms fail and how to distinguish meaningful diagnostic tools from misguided prediction products. Congress renews its push to bloc

In late October, the Roux Institute hosted Foundations for the Digital Commons with Bernstein Shur and RadicalxChange, convening technologists, policymakers, civic innovators, journalists, and funders to chart a path for renewing the digital systems that underpin democratic life. Over two days, participants examined practical models, from information flows and large-scale deliberation to data governance, while outlining actionable pathways for building modern digital infrastructure. Held in Maine, a state defined by democratic reforms and pragmatic problem-solving, the gathering reinforced a central insight that institutional design, more than technology, will determine whether digital innovation ultimately strengthens public trust.

AI agents are beginning to make market-shaping decisions. Hadfield and Koh’s new study reveals why this shift is significant. Current agents do not reason like economists, do not reflect human preferences, and do not fit into accountability structures. As governments experiment with agents in benefits, procurement, and infrastructure, Elana Banin reflects on the policy challenge of re-designing the rules before agents erode the market foundations on which democratic governance relies.

In the news this week, we examine the U.S. Air Force's plan to lease 3,000+ acres across five active bases to private data centers, raising questions about public oversight, security, and who benefits

The Air Force is quietly auctioning off slices of its bases for private AI data centers. They call it innovation; it looks like privatization. Fifty-year leases, 3,000 acres of military land, and no public say. If this is how we build the future, who’s really in command?

When refugees needed language support, Tarjimly turned everyday volunteers into lifelines. In this reflection, CEO Atif Javed traces how the platform evolved from a Facebook Messenger experiment into a global translation network, one now partially powered by AI. His key lesson is that designing for trust means using technology to amplify, not replace, human empathy in moments of crisis.

Brazil is pioneering AI-powered participatory governance at scale, processing input from 1.4 million citizens and turning public proposals into policy reports. From Maine’s 33 recommendations for resp

The Agentic State is an ambitious and inspiring blueprint for rebuilding government around AI agents that can act and decide autonomously. It powerfully diagnoses real failures in how the public sector designs, delivers, and manages services. While AI is giving us ways to accelerate change, the prescribed cure may be premature: most of what’s broken in government requires organizational reform, not automation.

The danger isn’t that AI will make us dumber—it’s that governments, companies, and schools won’t make us smarter with it. As policymakers stall and corporations automate, the burden of using AI wisely now falls on us. Like shoppers at the fish counter, we need to learn to read the labels—to know what’s safe, what’s risky, and how to choose well.

As headlines warn of “superintelligent AI” threatening human extinction, a new open letter reignites familiar fears. But beneath the apocalyptic rhetoric lies a deeper problem. The narrative around artificial superintelligence, long embraced by Big Tech, diverts attention from the real and immediate challenges of AI and how our democratic institutions can address them.

Hamburg uses AI to turn citizen feedback into urban policy, Boston showcases tools that improve public trust, and tribal nations are piloting AI for governance while protecting sovereignty. New York b

AI’s biggest breakthroughs were built on public datasets; the next wave should be, too. If governments make data AI-ready and keep access open, we can power tools that explain laws, improve services, and widen participation. In return, companies that train on taxpayer data should give back, through open licenses, benchmarks, and capacity that strengthen public institutions.

Explore, participate and contribute to the first collaborative public design map in the region.