Attributing the responsibility for AI's failures

What Ethics Can Say on Artificial Intelligence: Insights from a Systematic Literature Review

Giarmoleo et al. (2024)

Source DOI

Sub-category

Risk Domain

7AI System Safety, Failures & Limitations

7.4Lack of transparency or interpretability

Challenges in understanding or explaining the decision-making processes of AI systems, which can lead to mistrust, difficulty in enforcing compliance standards or holding relevant actors accountable for harms, and the inability to identify and correct errors.

"This section, constituting almost 8% of the articles, addresses the implications arising from AI acting and learning without direct human supervision, encompassing two main issues: a responsibility gap and AI's moral status."(p. 16)

Entity— Who or what caused the harm

Human

Due to a decision or action made by humans

AI system

Due to a decision or action made by an AI system

Other

Due to some other reason or is ambiguous

Intent— Whether the harm was intentional or accidental

Intentional

Due to an expected outcome from pursuing a goal

Unintentional

Due to an unexpected outcome from pursuing a goal

Other

Without clearly specifying the intentionality

Timing— Whether the risk is pre- or post-deployment

Pre-deployment

Occurring before the AI is deployed

Post-deployment

Occurring after the AI model has been trained and deployed

Other

Without a clearly specified time of occurrence

Supporting Evidence (2)

"AI moral agency and legal status (5.1%). This research line consists of two main issues. The first one concerns the existence and status of artificial moral agency (AMAs). Nowik (2021) analyzes the legal and ethical implications of attributing electronic personhood to AI in employment relations by looking at concepts like AI as an employer, liability, and mandatory insurance. Kornai (2014) discusses the moral obligations of autonomous artificial general intelligences (AGIs), as well as the challenges of bounding AGIs with ethical rationalism. Smith and Vickers (2021) examine how moral responsibility could be attributed to AI using a Strawsonian account. Other researchers discuss the design of artificial moral agents. Mabaso (2021) discusses the use of exemplarism, an ethical theory, in building computationally rational AMAs. Gunkel (2014) advocates for including robots and AI in moral considerations and offers a critique of the limitations of current moral reasoning frameworks. Wallach (2010) stresses the need for a comprehensive model of moral decision-making in developing artificial moral agents, with a focus on mechanisms beyond traditional cognitive factors"(p. 16)

"Responsibility gap (2.7%). This research reflects on the concept of the responsibility gap in AI, where an AI agent's actions that cause harm can lack clear responsibility. Saunders and Locke (2020) draw parallels between ancient practices of casting lots and AI in business decisionmaking and how, in both cases, control and moral responsibility are relinquished. Johnson (2015) discusses the potential emergence of a responsibility gap autonomous artificial agents of the future, emphasizing that responsibility allocation depends on human choices more than technological complexity. Awad et al. (2019) explore moral dilemmas in self-driving cars and propose that addressing these dilemmas requires collective discussions and agreements on ethical AI principles. Other scholars address responsibility gaps in AI systems, such as Santoni de Sio and Mecacci (2021), who identify interconnected responsibility gaps in AI and propose designing socio-technical systems for “meaningful human control” to comprehensively address these gaps. Schuelke-Leech et al. (2019) examine unexpected differences in the language used in policy documents and discussions about responsibility for highly automated vehicles."(p. 17)

Part of Human-AI interaction