Post date: May 20, 2019 1:54:43 PM
Humane Algorithms
Nick Gray
12:30 pm, 18 June 2019
Risk Institute Seminar Room
Institute for Risk and Uncertainty
Chadwick Building, University of Liverpool
Lunch will be provided. For catering purposes, please register at https://forms.gle/eReiqHaiUBtj75pV8.
Abstract: Autonomous vehicles are just around the corner, and computers already make many of the important decisions in our daily lives. Yet algorithms can make serious errors and misjudgements. They have been accused of formalising racial bias and prejudice in social media and in the justice system itself. Facebook's algorithms seem to have swung recent elections. The computers on the Boeing 737 MAX overrode human decisions in the disasters in the Java Sea and in Ethiopia.
Can algorithms be designed to interact more smoothly with human users? What would a humane algorithm be? How should algorithms address diversity and uncertainty? How should they balance risks? Can algorithms be designed to recognise fairness in automated decision making? How could this even be assessed? Can they be designed to fail in ways that avoid catastrophic outcomes?
This thinkover will focus on developing a useful definition for a humane algorithm, and a series of check points that will implement and make meaningful the definition in practical applications. The web site (https://sites.google.com/site/humanealgorithms/) will host the on-going collaboration. The work will go into a conference presentation.
All interested students and faculty members from all institutions, business people and industrialists are welcome to attend.