Duolingo English Test Responsible AI Standards

Abstract

Artificial intelligence (AI) is now instantiated in digital learning and assessment platforms. Many sectors, including tech, government, legal, and military sectors, now have used formalized principles to develop responsible AI standards. While there is a substantial literature around responsible AI more generally (e.g., Fjeld et al., 2020; Gianni et al., 2022; and, NIST, 20231 ), traditional validity frameworks (such as, Xi, 2010a; Chapelle et al., 2008; Kunnan, 2000; and, Kane, 1992) pre-date AI advances, and do not provide formal standards for the use of AI in assessment. The AERA/APA/NCME Standards (2014) pre-date modern AI advances, and include limited discussion about the use of AI and technology in educational measurement. Some research discusses AI application in terms of validity (such as Huggins-Manley et al., 2022, Williamson et al., 2012, and Xi, 2010b). In earlier work, Aiken and Epstein (2000) discuss ethical considerations for AI in education. More recently, Dignum (2021) proposed a high-level vision for responsible AI for education, and Dieterle et al (2022) and OECD (2023) discuss guidelines and issues associated with AI in testing. The Duolingo English Test (DET)’s Responsible AI Standards were informed by the ATP (2021) and ITC-ATP (2022) guidelines, which provide comprehensive and relevant guidelines about AI and technology use for assessment. New guidelines for responsible AI are continually being developed (Department for Science, Technology & Innovation, 2023).

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 16 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.