In the third session of TeachECONference 2024, we explored the transformative potential of Artificial Intelligence (AI) in education. This session featured presentations from experts, who shared their innovative approaches and findings on how AI can enhance learning and teaching. Here’s a summary of the key points and insights from the session. 

Nazanin Khazra: Enhancing Courses with AI Tools 

Nazanin Khazra from the University of Toronto commenced the session with discussing how she uses ChatGPT and other AI platforms to improve her courses and promote critical thinking and problem-solving. She emphasized the importance of future-focused education and shared insights from her course, “Big Data Tools for Economists”, covering topics, such as data visualization, GIS mapping, web scraping, and machine learning. The method involved 20%-30% of individual assignments’ grade on reflection, wherein students were provided with feedback and they had to employ AI to make changes and respond to the feedback, thus, producing a new version of the assignment, which was graded. Nazanin’s experiment underscored how ChatGPT helps students generate research ideas and think critically about potential challenges.  

Theodore Svoronos: Custom Chatbots for Enhanced Learning 

Theodore Svoronos from Harvard Kennedy School explored the use of custom chatbots to provide personalized learning experiences. He highlighted the potential of custom chatbots built upon ChatGPT to assist students in various tasks, from solidifying core concepts to generating unlimited practice problems that can be responsive to feedback. He also designed a virtual Teaching Assistant, by integrating rubrics and marking criteria, which helped grade students’ assignments. These methods provided transparency into student interactions with chatbots, allowing educators to monitor and improve the learning experience. 

Christian Tode: Evaluating AI’s Accuracy in Economics Education 

Christian Tode from the University of Applied Sciences Bonn-Rhein-Sieg presented their research on the accuracy and quality of ChatGPT’s (3.5 and 4.0) responses to economics concepts and multiple-choice questions, using the Core Econ “The Economy 1.0” as their reference textbook. Their findings accentuated that GPT-4.0 performed better than GPT-3.5, with fewer errors (3% versus 17%) in answering questions and providing explanations. They also found that explanations by both models cover most of the requested economic concepts, but often lack details and nuance. Additionally, he also emphasized that both models struggle with a few common errors, including factual inaccuracies and comprehension issues, particularly with less frequently used concepts. Thus, making the question/request clear and easy to understand by GPT is essential to accurately leverage this method.  

This blog was produced with the help of Microsoft Copilot.