Algorithmic Accountability: Drivers and Consequences of Software Developers’ Accountability Perceptions

Driven by rapid technological progress, algorithmic systems based on artificial intelligence (AI) are no longer used exclusively as supportive tools for decision-making but are increasingly embedded in daily tasks, resulting in a growing dependence of users and societies on them. Today, AI systems are applied in diverse domains such as finance, criminology, and credit allocation. Despite their widespread adoption, AI systems often exhibit serious misbehaviors that raise concerns about the ethical use of AI systems. Instances of systematic discrimination and harm to users highlight the urgent need for fairness, transparency, and accountability in the design and development of AI systems. Given the unacceptable and ethically indefensible nature of such situations, research and society are calling for greater fairness, transparency, and accountability in the design and development of AI systems. While the first two demands for more fairness and transparency have already been the subject of intensive research, the demand for more accountability has so far received little attention in information systems (IS) research. Against this backdrop, this dissertation investigates accountability in the AI context and examines how it can foster the design and development of more ethical AI systems by reducing the likelihood of harmful and discriminatory outcomes. For this examination, this dissertation adopts a practical-oriented approach and concentrates on perceived algorithmic accountability. Perceived algorithmic accountability refers to the awareness that an actor must explain and justify their actions and decisions related to an algorithmic system in front of a forum. Since numerous actors are involved throughout the lifecycle of AI systems, this dissertation focuses specifically on developers. This focus was chosen since developers are often associated with bearing accountability for AI systems, as they possess the technical knowledge to design and develop such systems. This capability enables them to understand and comprehend the actions and decisions of their AI systems, thereby allowing them to explain and justify the design and development of their AI systems. To achieve a comprehensive examination of developers’ perceived algorithmic accountability, this dissertation investigates both the drivers and consequences of developers’ perceived algorithmic accountability. Five studies employing diverse research methods reveal that organizational , project , and individual-related factors lead developers to perceive themselves as accountable for their AI systems. The findings further demonstrate that perceived algorithmic accountability yields both positive and negative consequences. On the positive side, it promotes code quality, more cautious and risk-averse design of AI systems, and demands for greater transparency and clearer communication within AI development processes. However, and on the negative side, heightened algorithmic accountability perceptions may also hinder innovation of AI systems by encouraging overly risk-averse AI design decisions. Through these findings on the drivers and consequences of developers’ perceived algorithmic accountability, this dissertation makes valuable contributions to IS research, organizations, and policymakers. First, the comprehensive examination of developers’ perceived algorithmic accountability allows IS research to better understand how developers behave within AI development processes and why they react to situations in these processes. This understanding is crucial for IS research in order to respond to developers’ resulting behavior before taking actions and decisions. Second, this dissertation is especially valuable for organizations, as it helps them understand why developers perceive themselves as accountable for their AI systems and perform actions and decisions. This insight is essential for organizations to assess the implications of perceived algorithmic accountability among their developers and to implement measures at an early stage to increase or reduce such perceptions. Thus, this dissertation makes an important practical contribution to making algorithmic accountability perceptions more controllable and influenceable. Finally, the findings of this dissertation provide policymakers with insights into the associated effects of algorithmic accountability perceptions, enabling them to make informed decisions when integrating algorithmic accountability into new or existing regulations. The identification of the associated effects of developers’ perceived algorithmic accountability equips policymakers with the opportunity to incorporate perceived algorithmic accountability into legislation in a deliberate and balanced manner without disproportionately burdening organizations and societies with the imposition of algorithmic accountability. In conclusion, this dissertation provides a basis for a better understanding of developers’ perceived algorithmic accountability. As exemplified in this dissertation, perceived algorithmic accountability contributes to AI systems being designed and developed more responsibly, which is important to reduce, or ideally eliminate, systematic discrimination and harm to users and societies and to enable the ethical and responsible use of AI systems.