Welcome to Frontiers - a series where we bring top researchers, engineers, designers, and leaders working at the cutting edge of various fields to go deep on their work with the Manifold Community.
For this talk, our speaker was Karan Taneja. Karan is a fourth-year Ph.D. student in Computer Science at the Georgia Institute of Technology, where he works under the guidance of Prof. Ashok Goel, director of the Design & Intelligence Lab and the National AI Institute for Adult Learning and Online Education (AI-ALOE). His research focuses on developing AI tools for education, leveraging Natural Language Processing and Knowledge-based Intelligent Systems.
Learn more about Karan's work on his website:
https://krntneja.github.io/
Abstract
As conversational AI becomes more prevalent in our lives, including our learning environments, ensuring its effective usage in educational contexts is of paramount importance. In this talk, I will be presenting my work on designing Jill Watson, a document-grounded virtual teaching assistant based on large language models (LLMs) for classrooms with diverse populations, including adult learners with unique preferences and learning needs. This AI system is designed to minimize hallucinations and ensure safe usage by students in college classrooms. We will discuss its evaluation in terms of response quality, safety, usage, impact on student performance, and students' perceived satisfaction. I will share our ongoing work in designing a multimodal Jill Watson that can address students' questions with text and figures grounded in course documents, presented in a blog-post-like format. This new system uses interactivity and engagement to promote learning and retention as well as explanations to promote trustworthiness. Finally, I will also introduce a method we developed to improve LLM-based AI agents like Jill Watson using a small amount of human feedback. We will present a novel method for active label correction, called ALC3, which uses filtering and auto-correction to maximize the utility of human feedback, and examine its performance on discriminative tasks common to conversational AI agents. My work spans natural language processing, human-computer interaction, learning sciences, and educational technology to take an interdisciplinary approach to bring conversational AI into mainstream usage in higher education.
We’re growing our core team and pursuing new projects. If you’re interested in working together, see our website for active initiatives and open positions, join the conversation on Discord and check out our Github.
If you want to see more of our updates as we work to explore and advance the field of Intelligent Systems, follow us on Twitter and Linkedin!
Written by
More from Manifold Research Group