City governments work hard to respond promptly to their residents’ needs. But despite their best intentions, rules and requirements often slow down processes and restrict how fast and responsive government workers be on their own.

Artificial intelligence may offer a fix for that challenge—the technology can help provide an avenue for city workers to use their judgment to deal promptly with an issue while adhering to rules in a transparent manner.

Stephen Goldsmith, the Derek Bok Professor of the Practice of Urban Policy at HKS, and coauthor Juncheng “Tony” Yang outline how AI can give government workers “accountable discretion” in an  article in Urban Governance. They write, “the paper introduces the concept of accountable discretion and proposes guiding principles, each linked to actionable measures: equal AI access, adaptive administrative structures, robust data governance, proactive human-led decision-making, and citizen-engaged oversight.”

We spoke with Goldsmith (who has also written about this work for Vital City) about why both discretion and accountability are important in city government and how AI can be used to balance these aims.

Q: This paper talks about “discretion” and “accountability.” Can you explain how you are thinking about those terms in the context of city government workers and the use of artificial intelligence?

Discretion has always been central to effective city government. In a phrase first coined by Michael Lipsky and HKS, street-level bureaucrats or frontline employees—inspectors, permit reviewers, social workers, police officers—constantly encounter situations where rigid rules alone are insufficient. They need judgment. Accountability is the democratic counterbalance: the obligation to explain and justify those decisions to supervisors, elected officials, and the public.

For decades, governments responded to fears of corruption or inconsistency by layering on rules and oversight. The result was often rigidity. Workers lost the flexibility to solve obvious neighborhood problems pragmatically.

AI changes how discretion is exercised and supported. AI can automate repetitive work, synthesize information, and surface patterns humans might miss. That allows employees to spend less time processing paperwork and more time solving problems. At the same time, AI creates new forms of accountability because decisions become more transparent, traceable, and reviewable.

So the question now is how cities design systems where AI enhances human judgment while preserving democratic oversight and public trust.

Q: You introduce the idea of “accountable discretion.” What does that mean in practice?

“Accountable discretion” means giving frontline workers more room to apply common sense and professional judgment while using technology to ensure decisions remain visible, reviewable, and fair. Traditionally, public administration treated discretion and accountability as a tradeoff: the more flexibility employees had, the harder it became to ensure consistency and oversight.

AI changes that equation. Accountable discretion means that public employees have greater authority to solve problems. A traffic engineer, for example, could instantly pull pedestrian counts, crash data, and neighborhood demographics before adjusting a walk signal rather than waiting months for a procedural revision. A health inspector could distinguish between an honest first-time mistake and a pattern of unsafe behavior.

At the same time, supervisors gain better oversight because AI systems create clear records of who made decisions, when, and under what circumstances. So the frontline worker spends less time typing and more time judging—but every judgment remains reviewable by AI-capable supervisors who review interpretations, ethics, and exceptions.

Q: Where are you seeing cities get this right?

The most promising examples are cities that use digital infrastructure and AI not simply to automate bureaucracy, but to make government more adaptive, responsive, and understandable to residents.

Portrait of Stephen Goldsmith.
“‘Accountable discretion’ means giving frontline workers more room to apply common sense and professional judgment while using technology to ensure decisions remain visible, reviewable, and fair.”
Stephen Goldsmith

Boston invested heavily in digitizing and visualizing curbside regulations and integrating open data into resident communication and operational management, rather than relying on static rules or siloed agency decisions. Santi Garces, Boston’s chief innovation officer, has enabled employees to access data more easily using AI tools. Portland has approached this as a change-management problem rather than purely a technology problem. The city has used targeted pilots and data-informed regulatory changes to influence behavior and build public support over time. What these cities share is a willingness to rethink operations around outcomes and responsiveness.

Q: You write that AI doesn’t just automate decisions but redistributes them. What does that actually look like in a city agency?

AI redistributes some of that authority across managers, technical systems, analysts, and even residents themselves. Take inspections. Previously, a supervisor could realistically review only a small sample of cases. With AI, supervisors can analyze patterns across entire teams. If one inspector consistently issues warnings where others issue fines—or if enforcement patterns correlate suspiciously with race or geography—the system can surface that immediately.

Meanwhile, frontline workers gain access to richer contextual information at the point of decision. AI instantly brings relevant records, histories, and comparisons into the workflow.

So discretion doesn’t disappear. It moves. Some authority shifts upward into oversight functions, some into technical system design, and some remain with frontline professionals handling contextual judgment and exceptions.

Q: Are current oversight systems, such as audits, regulations, and public input, equipped to deal with AI?

Not fully. Most oversight systems were built for traditional bureaucracies, not hybrid human-machine systems. Cities need new governance mechanisms: algorithmic impact assessments, continuous auditing, stronger data governance, and clear escalation procedures that allow employees and residents to challenge decisions when necessary.

But there’s another important point: AI oversight cannot become surveillance. Accountable discretion only works if governments establish clear limits on monitoring, maintain meaningful human oversight, invest in workforce training, and preserve straightforward ways for residents to question decisions.

Q: One of your arguments is that AI could increase citizen participation. How so?

AI lowers the barrier to understanding and interacting with the government. Today, many residents experience government as fragmented, confusing, and slow. AI can translate bureaucratic complexity into conversational interaction, helping residents participate more meaningfully in identifying problems and shaping responses.

Residents routinely surface issues on social media long before formal complaint systems capture them. AI tools can help governments detect neighborhood concerns earlier through sentiment analysis, survey tools, and sensor monitoring. That allows governments to become more proactive rather than purely reactive. AI also dramatically expands self-service and greatly broadens community leaders’ ability to make spatially organized natural-language requests of layered city data to understand issues and causes. Done well, these systems can make government feel more accessible, responsive, and understandable.

Q: Does anything worry you about how cities are currently adopting AI?

Yes. My concern is that many governments still approach AI primarily as a procurement issue rather than an organizational transformation challenge.

Technology alone will not modernize government. Cities have to rethink workflows, management structures, workforce training, labor relationships, and service design. Otherwise, they risk layering AI onto outdated bureaucracy. And governments need to address public anxiety directly. Residents need confidence that AI is being used to expose bias and improve responsiveness, not entrench unfairness or expand unchecked surveillance.

Q: How do you think AI will change the day-to-day experience of interacting with government in the next 5 to 10 years?

We write about and manage projects dealing with the responsive city. Even in two years, if incorporated thoughtfully, AI will help city leaders more quickly and comprehensively meet residents’ expectations in ways that build trust. And in so doing, government will become more conversational, personalized, and proactive.

Residents will increasingly interact with governments through intelligent systems that guide them through services end-to-end rather than forcing them to navigate agency structures independently. Applying for permits, resolving citations, opening a business, or accessing benefits should become substantially easier and faster.

Internally, city employees will spend less time on repetitive administrative work and more time on judgment, coordination, and problem-solving. AI will help governments identify patterns and root causes across agencies, enabling earlier, more intelligent intervention.

But the deeper transformation is institutional. 

—
Photo by David Paul Morris/Getty Images