Towards Interactive Research Agents for Internet Incident Investigation
Yajie Zhou, Nengneng Yu, Zaoxing Liu
Abstract
Investigating Internet incidents involves significant human effort and is limited by the domain knowledge of network researchers and operators. In this paper, we propose to develop computational software agents based on emerging language models (e.g., GPT-4) that can simulate the behaviors of knowledgeable researchers to assist in investigating certain Internet incidents and understanding their impacts. Our agent training framework uses Auto-GPT as an autonomous interface to interact with GPT-4 and gain knowledge by memorizing related information retrieved from online resources. The agent uses the model to reason the investigation questions and continuously performs knowledge testing to see if the conclusion is sufficiently confident or more information is needed. In our preliminary experiment, we build an agent Bob, who studies the impact of solar superstorms on the Internet and draws conclusions similar to those from a recent SIGCOMM paper written by a knowledgeable researcher. We envision this as a first step toward developing a future highly knowledgeable Internet researcher simulacra.