As part of Carnegie Mellon University’s efforts to support and promote the application of generative AI to education, the university has launched a seed grant program for the research and development of generative AI-enabled educational tools. The aim of the program is to foster new research, test and deploy tools to enhance education at CMU, and position awardees to secure additional funding for furthering their work.

In response to the program’s inaugural call for proposals, 43 proposals were submitted by a total of 87 principle and co-principal investigators, representing all seven of CMU’s schools and colleges as well as the University Libraries(opens in new window). 

“We are thrilled that the seed grant program received such an enthusiastic response from our faculty members,” said Provost and Chief Academic Officer James H. Garrett Jr(opens in new window). “The impressive number of proposals truly underscores our community's authentic commitment to pioneering new generative AI applications that will transform education at Carnegie Mellon and beyond.”

The proposals were evaluated by a panel of reviewers with relevant knowledge and expertise from the School of Computer Science(opens in new window), Dietrich College of Humanities and Social Sciences(opens in new window), Heinz College of Information Systems and Public Policy(opens in new window), the College of Fine Arts(opens in new window) and the Provost’s Division(opens in new window). The following three proposals garnered the highest endorsement, and Garrett has approved funding for the projects.

AI-Enhanced Writing Studio for Writing in the Disciplines and Professions

Despite significant research over the past 60-plus years, there have been no scalable breakthrough solutions to help college graduates meet standards of written proficiency. Dietrich College’s Suguru Ishizaki(opens in new window), David Kaufer(opens in new window) and David Brown(opens in new window) seek to close this gap through the application of research-based principles derived from the literature on writing process and pedagogy.

Building on their preliminary work in AI-enhanced writing environments for student writers, they plan to use the funding to enhance the learner experience; implement features to support instructors; evaluate the overall effectiveness of the environment; and understand how students write with AI-enhanced tools.

The professors plan to test the writing environment in an introductory statistics course and a professional writing course. Workshops for CMU instructors who assign writing in their classes are also planned.

MuFIN: A Framework for Automating Multimodal Feedback Generation Using Generative Artificial Intelligence

Feedback plays a critical role in improving learning outcomes in educational and professional settings. Traditional feedback methods, primarily textual, have been extensively studied and applied to facilitate learning and performance. In comparison, multimodal feedback — which integrates textual, auditory and visual cues — promises a more engaging and effective learning experience because it leverages multiple sensory channels, better accommodates diverse learning preferences and aids in deeper information retention. 

While generative AI has been primarily harnessed to automate and enhance textual feedback, its potential in crafting multimodal feedback remains largely untapped. A study proposed by computer science and psychology professor Ken Koedinger(opens in new window) and Jionghao Lin(opens in new window) and Eason Chen(opens in new window) in SCS seeks to bridge this gap by investigating how generative AI techniques can be employed to produce effective and scalable multimodal feedback.

Virtual Voice Coach: Improving Prosody and Expression in Vocal Art

Voice coaches train singers to sound believable in multiple languages through collaborative work on prosody, the patterns of stress and sound in singing. Jocelyn Dueck(opens in new window) in CFA and Shuqi Dai(opens in new window) in SCS proposed to work with new generative AI technologies to build an application that identifies good prosody in singing.

Similar to learning French well before traveling to France, prosodic nuance is vital to the compelling delivery of words in performance, including the communication of emotion. Opera singers perform unamplified and perceive sound from within and therefore require feedback from trusted collaborators, something vocal coaches normally provide.

A tool with state-of-the-art generative singing synthesis techniques could personalize vocal coaching using the singer’s own voices and habits, providing feedback even beyond the ability of human tutors, and could be readily available to the student right inside the practice room.

“My colleagues and I look forward to supporting these awardees' research and also to enacting additional strategies to help advance the broader set of AI-related educational innovations being pursued at CMU,” said Marsha Lovett, vice provost for teaching and learning innovation

One such ongoing strategy is the Generative AI Teaching as Research(opens in new window) (GAITAR) initiative. More strategies are anticipated in the fall.