Hoover Institution (Stanford, CA) — A quarter of civil servants are already using generative AI for public-sector work tasks, and that number is expected to grow rapidly, according to a survey developed by Hoover research fellow Oliver Giesecke.
While it trails the adoption of AI in private-sector workplaces, the use of AI by the public sector is growing quickly. One out of four employees report the use of generative AI for work tasks, with one out of six employees saying that they use the tools at least once a week. With widespread adoption, organization-level policies for responsible use and workforce training programs will be necessary soon, Giesecke’s survey results suggest.
“The adoption of AI in the public sector is quickly gaining momentum. Adoption at this point outpaces institutional safeguards and available training options,” said Giesecke, who is also a member of Hoover’s State and Local Governance Initiative.
With more than one thousand civil servants participating across twelve states, the survey asked workers about their current AI use patterns and how they expect to use AI tools. Further, the survey covered the perceived risks associated with AI use and policies that provide guardrails for responsible use. Lastly, the survey asked about investment and training needs to adopt the new technology.
With generative AI showing promise for productivity gains in a whole host of functions relevant to both the private and public sectors, the potential for US state and local governments is significant.
Adoption and Impact on Work
One out of four employees report the use of generative AI for work tasks, with one out of six employees saying that they use the tools at least once a week.
While the adoption is slower than in the private sector—where two-thirds of workers report generative AI use as of early 2024—the use of AI in the public sector is expected to grow quickly. Almost half of the survey participants in the public sector expect to use generative AI one year from now.
The overall adoption conceals large heterogeneities in use across job functions. Higher education staff (71 percent) and faculty (60 percent) report the highest use of generative AI, while the judiciary, social workers, corrections officers, law enforcement, and property managers have not adopted AI systematically (13 percent to zero).
Heterogeneous adoption patterns are also observed at the individual level. Employees with more advanced degrees and higher incomes drove the first wave of AI adoption, which raises interesting questions around whether the technology will exacerbate inequalities in the workplace. Moreover, employees in more senior positions have adopted AI at the highest rate across the workforce; this potentially creates the ground for a leadership-driven AI proliferation rather than leadership slowing down the adoption.
The first wave of AI adoption was focused on the drafting and editing of written communication, documents, and informational and marketing content. Chatbots and conversational AI powered this first adoption wave, followed by voice or speech recognition tools.
The reported increases in productivity and improvements in quality for AI-assisted work tasks are promising. While about 7 percent of civil service respondents said using generative AI saved them two or more hours of working time each day, another 44 percent said AI saved them between zero and two working hours each day. Similarly, a large share of public-sector employees reports quality improvements with AI assistance, with 70 percent saying that their work quality is better or significantly better with the help of AI.
Looking forward, the use of AI is expected to grow quickly in the public sector, where 44 percent of respondents are likely or very likely to be using generative AI one year from now. While current AI adopters are almost certain to continue using AI in the future, non–AI adopters are still uncertain about the use of the new technology.
When asked about the overall impact that AI will have on employees’ work in the next five years, 41 percent of public sector respondents are optimistic that AI will have a positive impact, 24 percent do not expect any material change in their work, and 16 percent have a negative outlook. The response differs significantly among AI adopters and nonadopters. Among AI adopters, three out of four respondents are optimistic about the impact on their work, while uncertainty prevails among the nonadopters.
Concerns About AI Use
Unsurprisingly, given highly publicized chatbot malfunctions and AI hallucinations dominating news headlines over the past two years, respondents in the public sector expressed concerns similar to those in the private sector about generative AI use.
When asked about their concerns specifically, the risk of providing inaccurate information was the most often cited concern, alongside security, data privacy, and transparency of operations using generative AI.
While some respondents said they have policies in place at their organization to mitigate possible harms caused by using AI, it appears that a large swath of the public sector is still unprepared to confront these risks. Nine percent of respondents said their organization had no formal policies on the use of generative AI, and another 51 percent said they were unaware whether any such policies existed.
These results indicate a need for sustained effort by public-sector organizations to develop system-wide AI use policies and plans for how to safeguard the data of members of the public who interact with civil servants wielding AI applications.
A Significant Desire for Training
Training for generative AI use in the workplace is underway, though demand currently outpaces the supply of training options. More than 70 percent of public-sector workers surveyed said they had received some formal training on the use of generative AI tools in the workplace, ranging from online courses to in-person workshops or external courses. The current training effort shows some success, with 71 percent of respondents who undertook the training saying it was effective.
Among those in the public sector who have not yet undergone any training in the use of generative AI at work, 49 percent of respondents said they were interested in receiving some form of training.
Training is a fundamental component in the adoption of the new technology, as reskilling of the current workforce is the most likely scenario for generative AI’s impact on the labor market. While layoffs are a prominent concern among all workers and firms when discussing the impact of generative AI, headcount changes are unlikely in the short term, and reskilling of the current workforce is the most likely outcome. However, there is tremendous uncertainty about the labor market impact, with 49 percent of public sector respondents saying that they had no idea how introducing AI to the workplace would impact staffing levels.
The survey evidence suggests that future-proofing current staff to the new technology for both effective and responsible use is critical. The current adoption pattern rate suggests a faster technological shift than with the personal computer in the 1980s or the internet of the late 1990s.
