Project Details
Program
Computer Science
Field of Study
Computer Science
Division
Computer, Electrical and Mathematical Sciences and Engineering
Faculty Lab Link
Center Affiliation
Resilient Computing and Cybersecurity Center
Project Description
Large Language Models (LLMs) have become ubiquitous in contemporary applications. They are trained on extensive collections of human writings, ranging from books, papers, and news articles to conversations on social media platforms. This comprehensive approach enables the development of sophisticated tools capable of emulating human interactions with remarkable fidelity. However, it is important to recognize that LLMs might inherit and perpetuate the biases inherent in human communications.
This project represents a concerted effort to delve deeply into the multifaceted landscape of biases inherent in interactions with LLM agents. By examining various dimensions of biases, we aim to explore how these biases manifest within LLM-mediated interactions. Through this project, we would not only understand the complexities of bias within LLM interactions, but also explore the possibility to mitigate, neutralize, or rectify these biases. This will underline if LLMs are more inclined to change opinions than humans. This approach underscores our commitment to fostering fairness, equity, and inclusivity in the realm of LLM-driven communication and interaction, ultimately advancing the societal impact and ethical integrity of LLM technology.
About the Researcher
Roberto Di Pietro
Professor, Computer Science
Affiliations
Education Profile
- Post-doc at the National Research Council ('04-'06), Pisa-Italy
- Ph.D. in Computer Science ('04), University of Roma ""La Sapienza"", Italy.
- Specialization Diploma in Operations Research and Strategic Decisions ('03), University of Roma ""La Sapienza"", Italy.
- MS in informatics ('03), University of Pisa, Italy.
- MS in Computer Science ('94). University of Pisa, Italy.
Research Interests
Professor Roberto's objective is to achieve excellence in cybersecurity research addressing both fundamental and applied challenges in the field, as well as to have impact and to generate innovation. In particular, Professor Roberto's research interests lie in the domain of security and privacy for distributed systems, with a special focus on systems supporting critical infrastructures. He is also interested (among others) in data science, on-line social networks, and application of AI techniques to solve security and privacy issues in current and future systems.Selected Publications
- Gabriele Oligeri, Savio Sciancalepore, Simone Raponi, Roberto Di Pietro: PAST-AI: Physical-Layer Authentication of Satellite Transmitters via Deep Learning. IEEE Trans. Inf. Forensics Secur. 18: 274-289 (2023)
- Savio Sciancalepore, Pietro Tedeschi, Ahmed Aziz, Roberto Di Pietro: Auth-AIS: Secure, Flexible, and Backward-Compatible Authentication of Vessels AIS Broadcasts. IEEE Trans. Dependable Secur. Comput. 19(4): 2709-2726 (2022)
- Roberto Di Pietro, Simone Raponi, Maurantonio Caprolu, Stefano Cresci: New Dimensions of Information Warfare. Advances in Information Security 84, Springer 2021, ISBN 978-3-030-60617-6, pp. 1-226
- Pietro Tedeschi, Savio Sciancalepore, Roberto Di Pietro: ARID: Anonymous Remote IDentification of Unmanned Aerial Vehicles. ACSAC 2021: 207-218
- Andrea De Salve, Paolo Mori, Barbara Guidi, Laura Ricci, Roberto Di Pietro: Predicting Influential Users in Online Social Network Groups. ACM Trans. Knowl. Discov. Data 15(3): 35:1-35:50 (2021)
Desired Project Deliverables
The project anticipates two primary outcomes: the creation of a system capable of simulating interactions within an online platform using LLMs as agents, and the utilization of the system to gain knowledge on the biases exhibited by LLMs across various operational scenarios.
Recommended Student Background
MS in Computer Science/Computer Engineering or final year MS
Currently PhD student