AgentPeerTalk: Empowering Students through Agentic-AI-Driven Discernment of Bullying and Joking in Peer Interactions in Schools
Project Overview
This document examines the application of large language models (LLMs), particularly ChatGPT-4, in educational environments, emphasizing their role in distinguishing between bullying and joking in peer interactions. It highlights the effectiveness of these AI systems in offering contextually relevant support to vulnerable students, thereby enhancing their well-being. The study also addresses the ethical and legal considerations necessary for implementing AI interventions and acknowledges the challenges faced by LLMs, including biases inherent in training data and limitations in understanding context. Ultimately, the findings suggest that agentic AI can play a crucial role in effectively addressing bullying in schools and promoting a supportive educational atmosphere.
Key Applications
Agentic-AI-Driven discernment of bullying and joking
Context: School environment, targeting students experiencing bullying
Implementation: Using ChatGPT-4 and other LLMs to simulate an agentic approach for real-time support
Outcomes: Improved discernment of bullying vs joking; enhanced support for students; positive impact on mental health
Challenges: Variability in LLM performance; political overcorrectness; limitations in context understanding and bias in training data
Implementation Barriers
Technical Barrier
Limitations of LLM context windows and inability to process complex legislative information.
Proposed Solutions: Manual input of legislative and cultural context to improve LLM responses.
Bias and Ethical Barrier
Political overcorrectness and biases in LLMs leading to refusal to comment on certain topics.
Proposed Solutions: Continuous fine-tuning of LLMs to align with evolving cultural and ethical norms.
Project Team
Aditya Paul
Researcher
Chi Lok Yu
Researcher
Eva Adelina Susanto
Researcher
Nicholas Wai Long Lau
Researcher
Gwenyth Isobel Meadows
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Aditya Paul, Chi Lok Yu, Eva Adelina Susanto, Nicholas Wai Long Lau, Gwenyth Isobel Meadows
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