The contribution of Luhmann’s concept of reflexive expectation to the development of AI governance

Artificial Intelligence is already impacting society in multiple ways, from the extraction of resources to build the entire AI ecosystem to the deployment of its services in a variety of social systems. In recent years, discussions and drafts around AI regulation have been raised, mostly by OECD and European Union, with limited normative ethics plurality. From a global perspective, the moral consensus of what is right and wrong became a challenging goal, considering the diversity of contemporary ethical positions in the global community, such as care ethics from the feminist branch or the communality ethics from the global south. In this context, sociology has been suggested to address the question of social order in the era of AI from a contingency and an interdisciplinary approach. More precisely, the theory of social systems of German sociologist Niklas Luhmann has been individualised as the most suitable lens to understand the complexity of AI’s impact on society for placing communication at the centre of the discussion about the interaction between humans and machines. Following the question of how social order could emerge in the AI era, where a highly complex system needs to be regulated by several prerequisites, the emergent approach has been identified as the best solution to understand the conditions of social order. From an extensive Luhmann’s
theoretical review, it has been identified a relevant structure called expectation that deals with the temporal contingency of moral generalisations and therefore deserved further attention. Through a thematic analysis of Luhmann’s expectation index, it has been understood the necessity of a governance that anticipates AI’s expectation through its own kind, hence, through another AI system because only the AI system could fit the accelerated dynamics of its own structure, observing, and describing alternatives in a timely manner. Furthermore, it is proposed a novel model of reflexive expectation to solve the bias issue in the double contingency of social systems, been advised further application of the model to AI.