SBIR Phase I: Mia Learning Independent Reading Choice Support System

This Small Business Innovation Research Phase I project will build children's motivation to read and help them access books best suited to their individual interests, purposes, and abilities. Even though robust research demonstrates both intrinsic motivation to read and print book ownership strongly shape reading achievement, few existing educational software products address them. The project will develop a voice chatbot app elementary students will use in class to receive personalized book recommendations and coaching on choosing well. An associated book subscription service will allow kids to own books they choose, including in very low income schools through partnerships with non-profits. Together, these offerings will tap a combined U.S. children's book and literacy educational software market for grades 2-5 that tops $1.4 billion dollars annually. Unlike most other "personalized" or "adaptive" learning systems, the app will use machine learning and artificial intelligence to increase the agency of students and teachers. It will focus on helping students improve their ability to make their own choices rather than making those choices for them. The intellectual merit of this project lies in its innovative combination of a recommender system and a pedagogical agent to simultaneously assist students in completing an authentic task (choosing books to read independently) quickly and well while also teaching them to complete the task increasingly effectively and independently over multiple performances. The voice conversational interface will provide this combined task support and coaching through an emotionally engaging narrative experience accessible to struggling readers. The research will yield a field-tested prototype of the system, constructing a domain model, authoring conversational content, developing machine learning technology, and iteratively improving the system through usability and pilot testing in elementary school classrooms. Technical challenges include tuning automated voice recognition in naturalistic classroom environments, overcoming the cold start problem to generate high quality initial recommendations, and supporting acquisition of both cognitive and metacognitive skills within an ill-structured domain where measurement of successful performance has complex dependencies with student identity and social context.