AI Lund lunch seminar: Abortive accountability? Critical perspectives by developers designing applications of AI for clinical healthcare

Beskrivning

Abstract: Increasingly, artificial intelligence (AI) is being applied to critical sectors of society such as healthcare. Arguably, AI is thereby gaining the status of an ‘infrastructural technology’ (Crawford & Calo, 2016; Star & Bowker, 2010), deeply integrated into the decision-making processes of key institutions and the lives of citizens and patients. In the past few years, the risk of harms and misapplications of AI, not to mention enhanced information asymmetry (Burrell, 2016), has effected a tremendous number of suggestions for how to ensure accountability in sociotechnical systems, including ethical principles particularly (Jobin et. al., 2019). As noted by Larsson and Heintz (2020), these principles form a part of the work on building an effective AI governance infrastructure. Other such tools and mechanisms in a infrastructure working to produce accountability may include, for example, engineering principles, professional codes, regulations, audits, tests, and, not least, standards (Bowker & Star, 1999; Star & Bowker, 2020). In this talk, I will tell about a case study focusing on the encounter between AI developers and three mechanisms for accountability that generally are highlighted in the academic literature on how to keep AI systems and developers in check and sustain key values in society: 1. Ethical principles for the design and development of AI 2.Methods for explainable machine learning, and 3. Technical standards implemented through audits. In this case, we see that developers are highly motivated to ensure accountability of their systems and their design and development practices.

Period	2020 sep. 16
Typ av evenemang	Seminarium
Plats	Lund, SverigeVisa på karta
Omfattning	Internationell