The GovLab: “…we are launching the Observatory of Public Sector AI, a research initiative of InnovateUS, and a project of The Governance Lab(opens in new window). With data from more than 150,000 public servants, the Observatory represents one of the most comprehensive empirical efforts to date to understand awareness, attitudes, and adoption of AI as well as the impact of AI on work and workers.
Our goal is not simply to document learning, but to translate these insights into a clearer understanding of which investments in upskilling lead to better services, more effective policies, and stronger government capacity.
Our core hypothesis is straightforward: the right investments in public sector human capital can produce measurable improvements in government capability and performance, and ultimately better outcomes for residents. Skill-building is not peripheral to how the government works. It is central to creating institutions that are more effective, more responsive, and better equipped to deliver public value.
We are currently cleaning, analyzing, and expanding this dataset and will publish the Observatory’s first research report later this spring.
The Research Agenda
The Observatory is organized around a set of interconnected research questions that trace the full pathway from learning to impact.
Our goal is not simply to document learning, but to translate these insights into a clearer understanding of which investments in upskilling lead to better services, more effective policies, and stronger government capacity.
We begin with baseline capacity, mapping where public servants start across core AI competencies, identifying where skill gaps are largest, and distinguishing individual limitations from structural constraints such as unclear policies or restricted access to tools.
We then examine task-level use, documenting what public servants are actually doing with AI.
Our data also surface organizational obstacles that shape adoption far more than skill alone. Across agencies, respondents cite inconsistent guidance, uncertainty about permissions, and limited access as primary barriers.
Through matched pre- and post-training assessments, we measure gains in technical proficiency, confidence, and ethical reasoning. We plan to track persistence through three to six-month follow-ups to assess whether skills endure, reshape workflows, and diffuse across teams.
We analyze how training shifts confidence and perceived value, both of which are essential precursors to behavior change. We collect indicators of effectiveness through self-reported workflow improvements that can later be paired with administrative performance data.
Finally, we examine variation across roles, agencies, and geographies, how workers exercise judgment when evaluating accuracy, bias, and reliability in AI outputs, and how different training modalities compare in producing durable learning outcomes…(More)”
