Online Program

Trustworthy AI in Practice (309957)

Keywords: Trustworthy AI, Data Science, Ethical AI

There are many examples of AI models that penalize women and minorities. In 2018, Amazon’s recruiting algorithm was found guilty of penalizing women’s application; in 2020, Apple’s credit card gave Steve Wozniak a credit limit ten times higher than his wife’s, despite the couple sharing all their assets. To deal with this problem, on one hand extensive work has been done to define AI ethical guidelines and regulations for a Trustworthy AI (for example, the Proposal for a Regulation on a European approach for Artificial Intelligence of the European Commission), on the other hand technical research on how to detect and mitigate bias in AI models. However, a gap still exists between theoretical ethical guidelines and regulations and practical application and implementation. In order to close this gap we need to build a data science process that incorporates social and technical aspects, with meaningful collaboration of societal stakeholders and AI developers. In this talk, we will bring forward some ideas on how to build such a data science process, and practical examples based on use cases.