Calls Grow for Culturally Responsive AI in Education

- Automatic Speech Recognition (ASR) systems struggle with young children due to fundamental differences from adult speech, including higher pitch, smaller vocal tracts, and greater variability in pronunciation and articulation. This results in a significant performance gap, as most ASR models are trained on adult speech data, and there is a scarcity of comprehensive, diverse datasets for children. - Reinforcement Learning (RL) is used to create adaptive content sequencing policies in educational tools. These models can optimize the sequence of learning activities, such as practice questions, to maximize a student's expected performance and learning outcomes based on their real-time progress. - Knowledge Tracing (KT) models, such as Bayesian Knowledge Tracing (BKT) and more recent deep learning approaches like DKT, are used to infer a student's evolving mastery of a concept. These models analyze a student's performance history to predict their future success on new tasks, forming a core component of adaptive learning systems. - To balance recommending known effective content (exploitation) with introducing new material to learn its effectiveness (exploration), some platforms model content recommendation as a multi-armed bandit problem. Contextual bandit algorithms, in particular, use a student's prior knowledge state as the context to inform which "arm" (or learning action) to pull next. - A study of the AI reading tutor Amira showed small but statistically significant positive effects on literacy scores for students in K-3rd grade. Other AI-powered tools like Plabook use speech AI to automate reading assessments, including oral fluency and phonemic awareness, reducing manual evaluation time for teachers from hours to minutes. - Designing AI for young learners involves robust safety protocols that go beyond typical applications. For children under 13, this includes compliance with the Children's Online Privacy Protection Act (COPPA), which limits the collection of personal data without parental consent, and implementing strong content filters to prevent exposure to inappropriate material. - Applying a Culturally and Linguistically Relevant Pedagogy (CLRP) framework can help guide the use of generative AI to support multilingual learners. AI tools can be used to facilitate translanguaging—the practice of using multiple languages to learn—and help teachers adapt curricula to better reflect students' cultural backgrounds.