Skip to main content Skip to navigation

Agent Smith: Teaching Question Answering to Jill Watson

Project Overview

The document elaborates on the innovative use of generative AI in education through the development of Agent Smith, a machine teaching environment that streamlines the training of Jill Watson Q&A agents for online learning. This system aims to tackle prevalent challenges in online education, including low student retention and insufficient instructor engagement, by allowing virtual assistants to efficiently respond to student inquiries. By significantly minimizing the time and effort needed to train these AI agents, Agent Smith has proven its effectiveness across diverse online courses at Georgia Tech. The implementation of this technology not only enhances the learning experience by providing timely support to students but also alleviates the burden on instructors, thereby improving overall educational outcomes. The findings suggest that the integration of generative AI like Agent Smith can play a critical role in transforming online education by fostering a more interactive and responsive learning environment.

Key Applications

Jill Watson Q&A system

Context: Online education, specifically in the Online Master of Science in Computer Science program at Georgia Tech.

Implementation: Using the Agent Smith machine teaching environment to train Jill Watson agents by mapping course domains to structured and unstructured knowledge bases and generating large datasets of question-intent pairs.

Outcomes: Significant reduction in training time from 500 hours to approximately 25 hours per agent, increased efficiency in answering student questions, and enhanced instructor presence.

Challenges: High initial time and labor investment required for training agents, reliance on domain experts for effective machine teaching.

Implementation Barriers

Cost and Expertise Barrier

Building AI agents can be costly and time-consuming, requiring significant labor from domain experts, who may also find it challenging to effectively teach machine learning systems due to the complexities involved.

Proposed Solutions: Utilization of machine teaching approaches to streamline the training process and reduce the time and labor required, alongside developing user-friendly interfaces and processes to decouple expert knowledge from machine learning system interactions.

Project Team

Ashok Goel

Researcher

Harshvardhan Sikka

Researcher

Eric Gregori

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Ashok Goel, Harshvardhan Sikka, Eric Gregori

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies