• Edit

Tsvi Kuflik

University of Haifa (Israel)

Tuesday July 13th, 2021, 15h30, Auditorium J. Herbrand, IRIT.

It Seems (Un)Fair to Me…Users' perception of algorithmic fairness: a framework and a case study

Abstract: Many algorithmic systems (AS) rely on Machine Learning (ML) and Artificial Intelligence (AI) for providing suggestions and recommendations to their users. The algorithms implemented in these systems are in most cases opaque and considered as “black boxes”. While these systems may provide accurate predictions, their reasoning process might be vague and problematic to convey to the general public who may operate the system or will be impacted by system's results. This opaqueness leaves cracks through which bias and discrimination can sneak in. Indeed, several such well-known cases have occurred in recent years, leading to the rise of the research area of algorithmic fairness that suggested numerous computational techniques for discrimination discovery and mitigating biases. Importantly, AS should not be examined merely on the merits of their computational fairness. Rather, the examination of peoples' perception of the systems' fairness is also called for. In this talk, an overall framework for algorithmic fairness will be presented, with a focus on the importance of users perception and the results of three online experiments, using a job recruitment system as a case study will be reported. The experiments' results suggest that the notions of individual and procedural fairness are perceived as the most important to the general public, while distributive and group fairness are the least important notions. The results of our study provide guidelines for creating, testing, implementing and explaining fairness of ASs that rely on ML and AI techniques.

Bio: Tsvi Kuflik is Full Professor of information systems, focusing on intelligent user interfaces. His BSc and MSc degrees are in Computer Science and PhD in Information Systems, all from Ben-Gurion University of the Negev. Since arriving at the University of Haifa, his research has focused on Intelligent User Interfaces — focusing on supporting individuals and groups in Ubiquitous Computing Environments and, more specifically, on problems related to Ubiquitous User Modeling — how to track, monitor and model users (and groups) in these environments and how to provide them with continuous personalized services based on their models. In this work, he aims are to suggest novel solutions for user modeling challenges in ubiquitous computing environments, by making user modeling a continuous, lifelong process, where the personal information can be made available to the environment as needed, under the control of the user. Since 2004 he leads a research group at the university of Haifa that continuous to work on exploring the potential of novel ICT to enhance the cultural heritage experience. In addition to this main research direction Tsvi Worked on agricultural decision support systems, where together with plant pathologists developed novel decision support systems for optimizing the use of pesticides. Recently he collaborated with a rehabilitation researcher on applying state of the art ICT for rehabilitation. In recent years he started to work on algorithmic transparency — making systems that apply Artificial Intelligence and Machine Learning techniques — more understandable to their users, where is this area he focusses on users perception of algorithmic fairness and the role of explanations in promoting trust in algorithmic systems.