Inhalt und Ziel des Forschungsprojekts
Aus soziologischer Perspektive sind die Zuschreibung von Entscheidungsfähigkeit und die damit zusammenhängenden moralischen Bewertungen empirische Fragen. In unserem Projekt interessiert uns deshalb, wie sehr das Verhalten einer KI in der Allgemeinbevölkerung als «Entscheidung» wahrgenommen wird und wie dies mit der moralischen Bewertung dieser neuen Technologien zusammenhängt. Unter welchen Bedingungen wird einer künstlichen Intelligenz Entscheidungs- und Verantwortungsfähigkeit zugeschrieben? Gibt es hierbei Unterschiede zwischen sozialen Gruppen? Dazu wird eine Kombination aus Experimenten, einer schweizweiten Befragung und qualitativen Interviews eingesetzt.
Wissenschaftlicher und gesellschaftlicher Kontext
Das Projekt leistet wichtige theoretische und empirische Beiträge zur Moral-, Kultur-, Technik- und allgemeinen Soziologie. Künstliche Intelligenz nimmt aber auch in der Wirtschaft, der Politik und dem persönlichen Leben einen grossen Platz ein. Entwicklung, Rechtsprechung, und Ethik sehen sich neuen Fragen gegenüber. Das Projekt hilft dabei, diesen neuen Herausforderungen soziologisch fundiert zu begegnen.
In contemporary societies, concepts of decision, decision-making, and choice are ubiquitous. Many human activities-from politics to love to art-are represented as the outcome of decision-making processes. But this tendency is particularly striking when we consider artificial intelligence (AI). The media speak of a self-driving car that “made a choice” to turn left or right, or we’re told about techniques to build artificial agents capable of “making complex decisions”. Artificial intelligence is a hot topic in public and scientific discourse; modern societies are struggling with this new technology, with its moral, legal, and societal implications. Importantly, these issues are often mediated by decision/choice concepts: AI systems should make “moral decisions,” should be “responsible for their choices,” or should be outright morally good agents.In a series of recent articles, Abend (2018a, 2018b, 2019) has argued that decisions/choice concepts play a decisive role in social practices and social organization. Social actors put decision/choice concepts to work. They are crucial for moral discourse, too. Decision/choice concepts are one aspect of the “moral background” (Abend 2014), a set of second-order elements that make moral life possible. When morality and decision/choice concepts meet, we speak of moral decisionism.Previous research in this area is scarce and has significant limitations. For one, there are no studies on the application of decision/choice concepts to AI and its relation to moral judgments. Second, previous research did not properly take the moral and social-structural background of social actors into account. Third, previous research did not systematically compare the effects of different determinants and analyze their interactions. Finally, quantitative, large-scale studies of the general population are missing. In short, a sociological approach to the study of the morality of AI is sorely needed.In our research project, we employ Abend’s conceptual framework, as laid out in his book, The Moral Background, and his articles on moral decisionism. Our two orienting research questions are as follows: (i) To what extent do social actors describe artificial intelligence with decision/choice concepts? (ii) To what extent do social actors describe artificial intelligence as a moral agent when doing X? We develop an explanatory model identifying situational and individual factors that account for the attribution of decision/choice concepts to AI, as well as moral agency to AI. Methodologically, we administer a factorial survey together with a questionnaire to a representative sample of the general Swiss population. Our research project will make several significant contributions to the scientific study of AI and its moral appraisal. We will contribute to various sociological subfields: sociology of decisionism, the sociology of morality, cultural sociology, sociology of science and technology, and sociological theory. Besides, we trust our findings will transcend the ivory tower. They can generate important insights for the design of “moral AI” and “explainable AI,” since being moral and explainable depend crucially on people’s beliefs about AI, for which a representative sample of the population will provide key information. Similarly, well-founded empirical results on decision/choice concepts in AI can inform public sphere discussions and policy-making. The existing experimental data is valuable, but we can benefit from a broader empirical picture: accurate knowledge of public perceptions will be an asset for policy aims, which must be sensitive to the publics and actors it is intended to serve.
Last updated:14.01.2022
