Artificial Intelligence in Education: Early Evidence and Open Questions

Generative artificial intelligence tools, including large language models that can produce text, solve problems, write code, and answer questions in natural language, entered educational settings at extraordinary speed following the public release of ChatGPT in November 2022. The arrival of widely accessible AI tools has forced rapid responses from schools, colleges, and universities that are struggling to determine appropriate policies for their use, what their effects on learning are, and how they should be integrated into pedagogy rather than simply restricted. The initial responses from educational institutions varied widely, from complete bans to enthusiastic integration, often within weeks of each other and sometimes reversing direction as experience accumulated. The practical limitations of detection tools, which cannot reliably identify AI-generated text, made enforcement of bans difficult and sometimes unfair, generating documented cases of false accusations against students. Many institutions have shifted from attempting to prohibit AI use to developing approaches that define when and how AI use is appropriate and that redesign assessments and learning activities for an environment where AI assistance is available. Research on the effects of AI tools on student learning is in its very early stages, limited by the recency of the technology and the challenge of conducting rigorous studies in educational settings that are changing rapidly. Available research and early evaluations document that AI tools can provide students with immediate feedback, help them structure their writing and thinking, and answer questions that might otherwise go unaddressed. These potential benefits are most significant for students who have limited access to tutoring and other learning supports. Research also documents risks and downsides of AI use in educational contexts. AI-generated text can be superficially persuasive while being factually incorrect, and students who accept AI output without critical evaluation may internalize misinformation. Research on student reliance on AI for writing finds evidence that students who use AI for drafting may develop less genuine writing skills than those who engage in more effortful writing processes. The trade-off between immediate efficiency and longer-term skill development is a central tension in education's response to AI tools. Academic integrity concerns are significant and have driven much of the policy response to AI in education. Assignments that ask students to demonstrate knowledge through writing or problem-solving can now be completed, partially or entirely, by AI with minimal student contribution. The implications for what assessments measure, and for how grades reflect student learning, are serious. Institutions and instructors who maintain assessment designs unchanged in the presence of AI tools may find that grades and credentials become less reliable signals of actual learning. Assessment redesign is the most significant pedagogical response to AI in education, and it requires substantial investment of instructor time and expertise. Assessments that require in-person demonstration of knowledge, application to specific and unfamiliar contexts, metacognitive reflection on learning process, or integration of personal experience are more resistant to AI substitution than standard essay prompts. Research on assessment innovation in response to AI is emerging and documents both promising approaches and practical challenges of implementation at scale. Equity concerns about AI in education run in multiple directions. Students with greater technological access, digital literacy, and AI sophistication may gain advantages over less privileged peers if AI tools enhance learning and performance. Conversely, institutions that restrict AI access or usage may disadvantage students who might benefit most from AI as a learning scaffold. Access to high-quality AI tutoring tools, which may require subscriptions or bandwidth, is not equitably distributed. Research on AI access and educational equity is an emerging priority. The role of AI in teacher professional development and in providing individualized learning support represents one of the most promising potential applications, with AI tools potentially helping teachers identify students who need additional support and providing students with personalized practice and feedback at scales that would not be possible with human tutoring alone. Early research on AI tutoring systems, including evaluations of Khanmigo and similar tools, is promising but preliminary, and questions about the quality of AI feedback and the conditions under which AI tutoring is most effective remain open.