AI Ethics - Algorithmic fairness and machine learning
The series' second speaker is Fredrik Johansson, Assistant professor in the Data Science and AI division, Department of Computer Science and Engineering, at Chalmers and the University of Gothenburg.
Software systems, and more recently AI systems, are used to in decision support with direct impact on human lives. This has raised concerns about the ethics and fairness of such systems: If a person applies for a loan, are they more likely to get it if they are male, despite controlling for income and background? If a risk scoring system is used to triage patients, will it prefer patients of certain ethnicities? These concerns have proven well-founded. In 2016, it was found that an AI system used for bail decisions in the USA was underestimating the probability of white criminals reoffending and overestimating the probability for black criminals.
In this seminar, we survey research that attempts to define algorithmic fairness and bias, as well as the means taken to improve this aspect of AI systems. We discuss the fundamental challenges of such an endeavour and the mathematical impossibility of a system that satisfies every kind of fairness.
Place: Scaniasalen, lecture room, Chalmersplatsen 1, Kårhuset
Welcome! (no registration required)