Biased mobile robots? Characterizing and addressing issues of fairness in mobile robotics

Add to your list(s) Download to your calendar using vCal

• Dr Martim Brandão (Post-Doctoral Research Associate, Department of Informatics, Kings College London)
• Wednesday 12 February 2020, 11:00-12:00
• Learning Centre, UG05.

If you have a question about this talk, please contact Hector Basevi.

Host: Dr Iran Mansouri (m.mansouri@bham.ac.uk)

Abstract: Machine learning has recently been the object of various criticism regarding issues of bias, fairness and discrimination. From criminal recidivism prediction, to facial analysis and natural-language-processing, many have been the examples of algorithms that perform worse on ethnic minorities, or otherwise reproduce social stereotypes. In this talk I will argue that similar issues of bias and fairness are present in the design of algorithms for robotics. I will focus on two case studies in mobile robotics: 1) pedestrian detection and 2) navigation/path planning. 1) Based on the analysis of state-of-the-art pedestrian algorithms I will show that there are currently some pedestrians that are more likely to be victims of collisions with robots. I will show the sources of this bias are many and relate to a complex socio-technical reality. 2) Through the use case of a rescue robot I will argue that robot navigation paths can indirectly discriminate, and I will formally define a few notions of fairness that could be used in motion planners directly. I will use these to characterize fairness, and to show the existence of counter-productive fairness definitions. I will conclude with some general insights for how we can correcly consider fairness when designing robots.

Website: https://www.martimbrandao.com/

This talk is part of the Artificial Intelligence and Natural Computation seminars series.

This talk is included in these lists:

• Artificial Intelligence and Natural Computation seminars
• Computer Science Departmental Series
• Computer Science Distinguished Seminars
• Learning Centre, UG05
• computer sience

Note that ex-directory lists are not shown.