AI as ‚virtual citizens‘ in a plural and dynamic society

Förderung durch die VW-Stiftung| Projektbeginn April 2019

AI are not just technical tools but rather 'virtual citizens', endowed with agency and discretion. Whereas their character as a black box is widely discussed, their moral inaccessibility is not. Broadening the issue of 'fairness' recently applied to algorithms, our research concern is as follows: How is the complex, typically ambiguous and contradictory moral infrastructure of late modern societies mirrored and eventually transformed by AI? In order to conceptualize this question, we combine the normative reasoning of the Sociology of Critique (SoC) resp. the Economy of Conventions with a coactive learning approach of AI/ML systems. SoC distinguishes several interwoven orders of justification, assuming that individuals have to justify their everyday doings and decisions with regard to competing moral orders. Fairness, transparency, accountability and trust in this view do not constitute fixed entities or system properties, but have to be continually negotiated and mediated referring to legitimate registers of justification. Transferring this approach to AI as 'virtual citizens' (1) learning machines will be confronted with conflicting moral orders and justifications. To ask how society is being translated into learning machines is to observe how plural orders of justification are actively adapted by AI systems. For this purpose, 'spaces of negotiation' will be created in which AI and humans interact based on a plurality of orders of worth.

Koordiniert durch das Lehrgebiet Soziologie II - Soziologische Gegenwartsdiagnosen


Professor Dr. Stefan Smolnik, Fernuniversitat Hagen,Fakultat fur Wirtschaftswissenschaften, BWL, insb.Betriebliche Anwendungssysteme

Dr. Carlos Castillo, Universitat Pompeu Fabra, Department of Information and Communication, Technology

Professor Dr. Kristian Kersting, TU Darmstadt, Computer Science Department, Center for Cognitive Science, Machine Learning Lab

Weitere Informationen finden Sie unter: