Responsible AI for Test Equity and Quality: The Duolingo English Test as a Case Study

AI in testing can improve efficiency but requires safeguards: The Duolingo English Test demonstrates that AI can streamline item generation and scoring of spoken and written responses, but these same capabilities introduce risks like bias in AI-generated content that must be actively managed through responsible AI practices.

Responsible AI practices directly support test fairness: The paper shows that implementing RAI standards—covering validity, reliability, fairness, privacy, security, and transparency—is essential for ensuring test equity across all test takers, particularly in high-stakes assessments where results affect educational and professional opportunities.

Transparency and accountability are critical for user trust: The case study emphasizes that AI-powered assessments must clearly communicate how AI is being used in testing, how decisions are made, and who is responsible for outcomes, which helps build confidence in the fairness of the assessment process.

Privacy and security protections are integral to responsible testing: The paper highlights that responsible AI practices must include safeguards for test taker data, particularly important given that language assessments may collect sensitive information about individuals' backgrounds and abilities.

Responsible AI requires intentional design from the start: Rather than treating ethics as an afterthought, the DET case study demonstrates that RAI principles should guide the development of AI-powered assessments from conception, including deliberate efforts to identify and mitigate potential sources of bias and unfairness.