This paper addresses two critical safety issues in conversational systems and methods to mitigate these problems. In section 1, I will discuss the problems faced by online conversational systems due to the actions of malicious users. It will particularly focus on the cyberpredator problem, which often targets vulnerable individuals, especially young children. In this section, I will review existing models to detect such predators, including my previous work, and present the results. In section 2, I will discuss safety issues related to conversational agents based on the Large Language Models. I highlight the limitations of existing works in assessing the safety of the models and propose a research topic that I plan to undertake to address them.
Paper
Full text
Safety Issues in Conversational Systems
OpenAlex · Hate Speech and Cyberbullying Detection · 2023
Abstract
This paper addresses two critical safety issues in conversational systems and methods to mitigate these problems. In section 1, I will discuss the problems faced by online conversational systems due to the actions of malicious users. It will particularly focus on the cyberpredator problem, which often targets vulnerable individuals, especially young children. In this section, I will review existing models to detect such predators, including my previous work, and present the results. In section 2, I will discuss safety issues related to conversational agents based on the Large Language Models. I highlight the limitations of existing works in assessing the safety of the models and propose a research topic that I plan to undertake to address them.