How AI Supports Fair and Valid Language Testing

As artificial intelligence becomes more visible in education and assessment, many educators and administrators have questions about how AI is actually used in language testing, and how concerns about fairness, bias, and validity of AI-enhanced language tests are addressed in real-world settings. This session provides a practical, non-technical overview of how AI supports test design, administration, and scoring in a high-stakes English proficiency assessment. Using the Duolingo English Test as a concrete case, we describe how AI tools assist with tasks such as generating test content, managing secure test delivery at scale, and supporting scoring processes. Rather than replacing human expertise, these systems are intentionally designed to operate within a human-in-the-loop model, where assessment specialists review AI-generated materials, monitor test performance, and intervene when anomalies or equity concerns arise. The session emphasizes process over theory, focusing on how responsible AI use is operationalized in practice. Topics include how test developers determine which components are appropriate for automation, how quality control and bias checks are embedded throughout the test lifecycle, and how accountability is maintained when automated systems are involved. The role of transparency—both for institutions using test scores and for test takers themselves—is also addressed. In addition, the session considers issues of access and inclusion, illustrating how technology can expand global testing opportunities while maintaining consistent standards. Participants will leave with a clearer understanding of how AI is currently used in cutting-edge language assessment, what safeguards are in place to protect test takers, and how educators and administrators can critically evaluate AI-assisted tests when making admissions, placement, or policy decisions.