Enhancing Educational Dynamics Integrating Large Language Models with a Social Robot

In contemporary educational settings, there is an increasing demand to enhance social interactions and learning opportunities using innovative technologies. Traditional robotic interactions often fall short of providing the responsiveness and personalization necessary for effective educational support. This research addresses these limitations by exploring the usage of Large Language Models (LLMs) to deliver more dynamic, responsive, and customized interactions in educational environments where robotic technologies are becoming more prevalent. An experimental approach is used to systematically evaluate different integration methodologies, ranging from basic LLM-enhanced interactions to more complex customized interactions. The primary objective of this study is to determine whether the integration of LLMs can significantly improve the interaction quality of Pepper the robot compared to traditional scripted methods. Findings show that LLM integration, particularly with cloud-based models like OpenAI, substantially improves interaction quality, response times, and user satisfaction. Local LLMs offer additional benefits in terms of data privacy and offline functionality, while the integration of could-based LLMs enhances user interaction and satisfaction, challenges such as cost, data privacy, and technological constraints must be addressed to fully unlock the potential of AI-enhanced social robotics in education.