Year: 2021
Type: dissertation
Abstract: The widespread application of AI comes with high expectations of benefits for many domains, including healthcare, finance, education, and transport. However, there have been equally alarming predictions of discriminatory systems, biased decision-making, and privacy violations. This has sparked a public concern and discussion on applied AI ethics. Prior research on AI ethics has been highly theoretical and conceptual, focusing on creating ethical AI guidelines and frameworks, whereas the empirical research is still under-researched and poorly available. Therefore, this study aims to fill this gap between theory and practice by conceptualizing some of today’s ethical AI practices. This study was set out to examine how ethical AI principles are put into practice in organizations developing or deploying AI, and to find out what are the drivers of implementing ethical AI. Overall, 13 semi-structured expert interviews were conducted with 12 AI organizations, and the data was analyzed following the Gioia method. As a result of the analysis process, it can be concluded that ethical AI principles are implemented as governance, and AI design and development practices, as well as knowledge management and stakeholder communication activities. As for the drivers, the ethical AI practices are motivated by trust and risks, regulatory and stakeholder pressure, and business drivers. Furthermore, the identified individual ethical AI practices and drivers are discussed in this research. The findings indicate that AI organizations had implemented relatively similar ethical AI practices that were recommended by today’s AI guidelines and frameworks. Moreover, ethical AI practices were not purely considered important for ethical reasons. Instead, organizations were more motivated by the pragmatic drivers, such as regulatory requirements, stakeholder pressure, maintaining customer trust or managing risks.
Source: UTUPub (University of Turku)
Author: Akseli Seppälä
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3152636309