U-GAIN Reading Overview and Demonstration

• In Event: Innovating K–12 Teaching and Learning with Generative AI: Insights From IES-funded National R&D Centers

Abstract

Purpose
Scores from the National Assessment of Educational Progress show continuing declines in reading performance, a problem that many educators find deeply troubling (Goldstein, 2025). As we address this challenge, we align with the United States Department of Education priority on evidence-based literacy, to "...promote literacy instruction based on evidence to ensure that proven methods based in the science of reading will be used to help students learn to read." U-GAIN Reading investigates how Generative AI can better implement the science of reading in ways that adapt to important variations in how students read, speak, and interact as they practice reading with a large-scale online reading tutor.


Theoretical Framework
U-GAIN Reading builds on the Active View of Reading (Duke & Cartwright, 2021), which goes beyond the Reading Rope (Scarborough, 2001). One important component of this view is "bridging processes" — components of reading that interconnect word recognition and language comprehension. Technologically, U-GAIN Reading builds on Project LISTEN, a long-term, empirically successful (Mostow, 2016) effort to use automated speech recognition to tutor students as they read aloud. U-GAIN Reading incorporates an Ethics team to ensure safety, fairness, and responsibility, e.g., the EdSAFE AI Benchmarks (EdSAFE AI, n.d.).

Demonstration Overview
This poster-demonstration will show both baseline and new prototype technologies. Our baseline demo will show Amira Learner's valid and reliable diagnostics for early reading, which form a measurement context for our research. The existing platform also identifies learner needs as they learn, provides tutorial guidance, and adapts to each learner. New prototype technologies will illustrate how GenAI is enabling adaptation to differences in how learners read aloud, how they engage with technology while reading, and how generating texts and dialogues can maintain cognitive engagement.

Methods and Data Sources
We will provide an overview. U-GAIN studies employ mixed methods across both small and large scale data sets. Classroom-level studies are exploring opportunities and barriers in usage of the technology, while our larger datasets allow for exploration of patterns in usage, engagement and learning and how these co-vary with students' linguistic backgrounds and school setting. Learning sciences co-design methods are combined with computer science and learning analytic methods.

Scientific and Practical Significance
Scientifically, we address three interrelated problems: (a) how to interconnect the potential of generative AI with the science of reading (b) how to adapt ASR technology to the range of ways striving readers speak as they read aloud and (c) how to use effectively use GenAI to tutor students who are still learning to read. Practically speaking, U-GAIN Reading has found very high interest among school leaders and edtech procurement teams in applying GenAI to accelerate how much striving readers learn through independent practice. Further, because we are conducting our research in partnership with a platform that serves millions of students every week, we have the opportunity to deliver the results from this research at scale—and we will also produce research publications, data sets and training that support the growth of research capacity in this area.

Authors

• Jeremy Roschelle, Digital Promise Global

• Adam Porsch, Amira