InnovateUS 
The public conversation about artificial intelligence has swung rapidly from exuberance to alarm and dire warnings about autonomous agents, cyberattacks, and even human extinction.
These risks require serious attention and strong governance. But a debate framed only around catastrophe casts AI as an autonomous force advancing along an inevitable path. It makes a handful of Silicon Valley executives the protagonists, whether as saviors or villains, while reducing governments and the public to spectators. That framing obscures a basic democratic fact: democratic institutions design, build, buy, regulate, and use AI, and societies still choose where it leads.
In 1966, the British architect Cedric Price gave a lecture titled “Technology is the answer, but what was the question?” Sixty years later, that question has become newly urgent for governments deciding what forms of AI to build, buy, regulate, or refuse.
Our goal at Reboot, and in the eponymous Reboot: AI and the Race to Save Democracy, is to ask: what do governments need to do differently in the age of AI to give people greater agency, expand meaningful participation, strengthen social cohesion, and govern more effectively?
Over the past year, our work (via our learning arm, InnovateUS) delivering AI learning to more than 250,000 public professionals has revealed that many are only beginning to use generative AI. Workers often do not know how AI applies to their work, whether they are permitted to use it, or how to use these varied tools well to deliver services, uphold rights, solve public problems, and improve people’s lives.
The next phase of public-sector AI must therefore begin with a harder question: What are we using AI for?
To advance that conversation, Reboot will focus this fall on four connected questions:
What capabilities must public institutions build?
How can they become institutions that listen and learn?
What public infrastructure do they need?
And how will they measure whether AI is creating public value and contributing to democratic renewal?
Building Institutional Capability
Governments need the leadership, management, product, procurement, and technical capacity to set a vision for why we adopt AI and how we can use AI to transform how our institutions work.
Building capability also means engaging workers in AI decisions from the outset. Frontline employees should help define the problem before selecting a tool, test how it changes their workflows, identify risks and unintended burdens, and help determine what success should look like.
Without a whole-of-workforce understanding of what AI is and is not, governments cannot be informed buyers, effective partners, or transformative users in the public interest.
This fall, we will ask and endeavor to answer:
What should every public professional understand about AI?
How should learning differ across roles and levels of responsibility?
What must leaders do to turn individual skills into organizational change?
Which capabilities must government retain in-house?
And how can procurement, product management, and worker-led innovation help institutions move from isolated experiments to sustained improvements in public service?
Without a baseline understanding of what AI is and is not across the workforce, governments cannot be informed buyers, effective partners, or transformative users in the public interest. Individual skills become institutional capability only when employees also have clear rules, access to appropriate tools, supportive leadership, and the authority and opportunity to apply what they know.
Creating Institutions That Listen and Learn
Digital technology has made it easier than ever for people to speak. Public institutions still struggle to process what they hear, learn from it, and connect it to consequential decisions.
Governments already receive enormous amounts of knowledge through consultations, complaints, audits, contact centers, public meetings, employee feedback, and everyday service delivery. Yet resident and frontline worker knowledge about which services are difficult to navigate is filed away and never reaches someone with the authority to act.
Institutions could make better decisions by drawing on the collective intelligence of the people they serve. Meaningful participation also gives people greater agency and shows that the government can respond.
AI could help institutions analyze more information, identify recurring problems, and surface areas of agreement as well as less visible perspectives.
But processing more input does not guarantee that an institution is listening. Automated summaries can erase disagreement, reproduce bias, or make it impossible to trace a conclusion back to what people actually said. Faster consultations can still be performative. Listening happens only when public knowledge can change what an institution understands and does.
This fall, we will ask:
Whom should institutions engage, what should they ask, and how should public input shape decisions?
How can AI help make sense of large volumes of information while preserving minority viewpoints and disagreement?
What standards of fidelity, plurality, traceability, and contestability should apply to AI-assisted synthesis?
What institutional processes can connect listening to action?
And how should governments determine whether engagement has strengthened public agency, trust, social cohesion, and the quality of their decisions?
The challenge is no longer simply getting people to speak. It is building institutions that can listen, learn, and act on what they hear.
Building Public Infrastructure
Governments are rapidly becoming dependent on AI systems they do not build, cannot inspect, and may struggle to replace. This fall, we continue exploring Public AI and how to ensure AI infrastructure operates in the public interest.
We will examine:
What this infrastructure looks like in practice, including efforts in Europe and elsewhere to develop shared models, knowledge resources, and public capabilities.
The physical infrastructure beneath AI. These systems rely on electricity, water, land, computing power, and public investment. Decisions about who supplies those resources, who pays for them, and who shares in the benefits are public choices.
How governments can build, buy, and govern AI systems so that public needs drive decisions, including the role of open standards, interoperability, model diversity, and procurement in reducing vendor lock-in.
How governments, universities, cultural institutions, civil society, and communities can collaborate on shared data, models, computing capacity, and reusable tools that no single institution could develop or sustain alone.
Building public infrastructure gives governments greater control over the technologies they use and greater leverage over the markets they help create.
Measuring Public Value
AI is often evaluated through model benchmarks or efficiency measures. Neither tells a government whether a particular product will work reliably in a real workflow or improve outcomes for the people it serves.
Evaluation must distinguish among the underlying model, the product built with it, and how that product is used. At each level, governments can make practical choices about what they build and buy, what evidence they require, and how they align AI with democratic values.
Drawing on our Practical Approaches to Evaluating AI for Public Benefit series, we will examine:
How governments can define public benefit with workers and residents and translate values such as quality, access, equity, dignity, trust, and responsiveness into practical measures.
What agencies can learn from evaluating the model, the product, and its use, and what they must test for themselves.
How governments can compare people, AI, and human-AI teams and monitor performance and unequal effects over time.
What evidence should lead an institution to adopt, scale, redesign, replace, or discontinue a system.
Evaluation is how a learning institution determines whether AI is actually improving government.
These four themes trace a connected path from purpose to practice.
Institutional capability gives governments the ability to articulate a vision, develop a strategy, and implement it.
Listening and learning describe how government must transform, continuously drawing on residents' knowledge, workers' expertise, and evidence to improve how it operates.
Public AI concerns the tools, data, computing capacity, standards, and partnerships governments need to build, buy, and share to support that transformation in the public interest.
Evaluation provides the evidence to determine what is working, for whom, and under what conditions, and whether an initiative should be scaled, redesigned, or stopped.
Over the coming months, Reboot will publish fewer, sharper, more argument-driven pieces. We will draw on our research and practical work, turn case studies into lessons that governments can use, and ask what practitioners should do differently as a result.
We will continue to take AI’s risks seriously. But seriousness cannot mean surrendering our capacity to act. Democratic societies must build institutions capable of governing the most dangerous forms of AI while making deliberate choices about where other forms might improve governance and deepen democracy.
The question for Reboot this fall is not simply whether government should use AI. It is what we want democratic institutions to become, what public purposes they should serve, and what role, if any, AI should play in getting us there.