Eli Orbach examines how the United States’s decentralized education system impedes the diffusion and adoption of generative artificial intelligence in K-12 schools. Slow and uneven diffusion will exacerbate current socioeconomic inequalities, harm students’ future work prospects, and impede macroeconomic growth and productivity.
The American K-12 education system is fragmented across 50 states and six territories, over 13,000 public school districts, and a range of private and semipublic institutions. This structure has long benefited students and the education sector, fostering variety among schools as well as innovation in school policy and pedagogy. Yet in periods of rapid technological development, decentralization can act as an impediment to technological diffusion, both slowing adoption and creating inequities. We find ourselves in exactly this scenario with generative artificial intelligence (GenAI).
The appropriate approach to implementing GenAI—whether to enhance productivity for teachers, tutor students with personalized lesson plans and feedback, or teach whole curricula—remains contested. Regardless, schools must adapt to the technology’s growing presence. Education industry icons such as Khan Academy founder and CEO Sal Khan have repeatedly espoused the pedagogical advantages of GenAI. Furthermore, according to an open letter signed by over 250 CEOs, “A basic foundation in computer science and AI is crucial for helping every student thrive in a technology-driven world.” Politicians and policymakers across the political spectrum agree. President Donald Trump wrote in a recent executive order that “We must provide our Nation’s youth with opportunities … to use and create the next generation of AI technology.” The Biden administration took a similar stance, as did the Department of Education in its May 2023 report on AI and the future of learning.
The nature of GenAI diffusion in our school systems will have social and macroeconomic implications stretching beyond just the education sector. An uneven rollout will exacerbate existing inequalities, as students without access to GenAI will lose out on an important skillset. Slow diffusion also threatens to weaken economic growth and delay the emergence of new complementary products or markets. In contrast, an effective rollout promises large productivity gains for future workers.
Despite the broad support for and importance of GenAI adoption in the education system, the technology has yet to be integrated into the vast majority of school curricula. Even schools with an articulated desire to forge ahead face significant barriers to GenAI adoption, including informational and skill-based barriers, ethical concerns, and supply-side limitations. The decentralized nature of the United States education system is another barrier that has received little attention. This article explores the structure and logic of decentralization in the U.S. education system, the collective action problem it poses for schools’ adoption of GenAI, and a realistic policy framework for coordination that respects local autonomy and the existing educational system.
The decentralized education system
The federal government plays a limited role in the K-12 education system—providing roughly eight percent of public school funding while generally confining its regulatory role to matters of social equity, civil rights protection, and broad oversight. The recent dismantling of the Department of Education (including mass layoffs, particularly in the Office for Civil Rights and the Federal Student Aid Office) further diminishes federal influence. Regulation mostly occurs at the state level, yet states likewise have limited authority. Local administrators and elected officials at the school and district levels manage day-to-day operations. Public schools face significantly greater levels of regulation than private schools, and charter schools fall somewhere in between.
Decentralization helps to create an exceedingly complex educational bureaucracy. The division of responsibility among states, districts, and individual schools varies. For example, 19 states have lists of approved textbooks for public school use, while the rest allow more freedom to local officials, schools, and teachers. Only infrequent federal mandates apply ubiquitously.
Education in the U.S. is decentralized for good reasons. By delegating power to local authorities who are in closer proximity to the communities they serve, schools are better able to provide for the needs and desires of students. Moreover, decentralization enables—in the famous words of Justice Louis Brandeis—“laboratories of democracy,” where states (as well as school districts and individual schools) can experiment with their operations and allow others to replicate their successes, thus promoting organic policy innovation at low risk to the broader system. Charter schools illustrate the virtue of laboratories: after Minnesota passed the first charter school law in 1991, the model’s compelling design and resonance among families and school organizers propelled eight other states to adopt similar laws within two years.
Decentralization also permits schools to reflect the diverse values and priorities of the local communities they serve. Followers of Catholicism, for example, value the existence of Catholic schools, and the absence of institutions with alternative religious affiliations would deprive others of the same benefit.
GenAI’s collective action problem
For all its merits, decentralized education engenders both inequities and delays in the diffusion of GenAI. Schools’ adoption of GenAI faces a collective action problem: the education system—and the wider economy—would benefit from rapid and equitable diffusion, yet achieving that outcome requires coordination.
Some degree of standardization across the education system is essential. Students leaving school—whether bound for the workforce or higher education—are expected to understand certain concepts and possess certain skills (e.g., a pre-med student must have basic proficiency in biology and chemistry). Similarly, students should be able to transfer between schools with minimal disruption. In the face of such constraints, educational institutions design compatible curricula. Students across the country take the SAT and other standardized tests, learn math through roughly the same sequence of courses, and write essays using the same five-paragraph format. Moreover, since custom educational resources are prohibitively costly for all but the wealthiest schools, most schools rely on standardized, off-the-shelf products like textbooks and pre-made worksheets.
The established, standardized structure is the path of least resistance for pedagogy. Schools attempting to deviate from the path—by deeply integrating GenAI, for instance—encounter higher costs; decreased compatibility with the programs of other schools, universities, and employers; and a scarcity of off-the-shelf products and services. But once a sufficient number of schools and districts adopt GenAI (when a “critical mass” is achieved), the forces of standardization will shift to accelerate, rather than impede, its diffusion. GenAI educational tools will be created to meet growing demand, and educational standards will evolve. This is not purely theoretical. For example, schools reached a critical mass in their adoption of the internet in the mid 1990s, and classroom access increased twentyfold by 2000. Subsequently, the internet has become central to education, streamlining communication and school organization.
Here is where the collective action problem arises: reaching critical mass requires a coordinated push on the part of many states, districts, or schools. With time, this will happen organically as school districts experiment with GenAI on their own, and especially as technological improvements and information gains bring down adoption costs. However, the organic, unassisted process of GenAI adoption is unlikely to proceed quickly or equitably.
The adoption process for GenAI is far from straightforward. Over the last couple of years, a large body of literature has developed concerning the proper adoption methods for GenAI. Internal arguments within schools and districts on the appropriate use-cases of GenAI have become commonplace. Can students use GenAI to draft or edit essays? Can teachers use GenAI to grade assignments? A recent poll from Gallup and the Walton Family Foundation shows that over a quarter of educators still fully object to the use of GenAI in schools—change often invites opposition. Given these hurdles, it is unlikely that a critical mass will agree on a unified approach in the near future.
Similarly problematic are the barriers that, for certain schools and districts, prevent GenAI adoption altogether. The use of GenAI requires high-speed internet connection, something a quarter of schools still lack. Furthermore, schools and districts serving disadvantaged communities have fewer resources at their disposal to invest in innovative GenAI testing and adoption. This not only creates further inequities but also shrinks the pool of potential GenAI innovators and thus induces more delays.
A modest policy recommendation
The current decentralized education system fails to promote the swift and equitable diffusion of GenAI. Some top-down coordination among states, districts, and schools is essential to achieve widespread adoption. However, any central organization should respect local and state autonomy and values. An effective policy framework would:
1. Create a standard framework for GenAI adoption. The federal government should—likely through an existing agency or administrator—create a national standard for GenAI implementation in schools. Since the best course for GenAI adoption is still hotly contested, the standard should offer a degree of flexibility and be developed in consultation with states, teachers, district officials, and GenAI experts. This could function similarly to the Common Core Standards Initiative, which aims to ensure that students receive adequate education in English and math. Note that Common Core was created by a large coalition of state leaders rather than the federal government, but the same principle applies.
2. Utilize incentives, not mandates. Schools, districts, and states should not be required to accept the standard framework, as a mandate would infringe upon their autonomy and invite numerous legal challenges. Rather, the federal government should incentivize the standard’s adoption through grant payments or informational campaigns.
3. Address inequities in education. For schools to adopt GenAI, they must have the capacity to do so in the first place. Expanding internet coverage and ensuring access to the necessary investment capital for schools in poorer areas should be a priority. Coordination cannot truly exist unless all players can participate.
Conclusion
The swift and equitable diffusion of GenAI is a national priority, no less in K-12 education. However, the decentralized nature of the U.S. educational system, in conjunction with the collective action problem inherent in the adoption of new educational technologies, results in slow and imbalanced adoption. Federal government policy that addresses the challenges in coordination while respecting the existing decentralized institutional structure is critical.
Author Disclosure: The author reports no conflicts of interest. You can read our disclosure policy here.
Articles represent the opinions of their writers, not necessarily those of the University of Chicago, the Booth School of Business, or its faculty.
Subscribe here for ProMarket‘s weekly newsletter, Special Interest, to stay up to date on ProMarket‘s coverage of the political economy and other content from the Stigler Center.
