The blame game

: attribution of responsibility in human, black box and explainable AI in the context of successful and unsuccessful managerial decision-making

Abstract

The rise of ChatGPT, deep fake artificial images, and automated machine learning techniques are proof of the growing demand for usable AI methods. The more sophisticated those applications become, the harder it is to create transparency along the responsibility chain. This master’s thesis looks into the complex world of responsibility attribution in collaborative human-AI decision-making with an emphasis on different types of AI and different decision outcomes within managerial contexts. The findings support a general trend: compared to AI entities, people tend to place more blame on human decision-makers, which is consistent with the fundamental attribution error. Contrary to predictions, the research finds no significant difference in the allocation of blame between explainable AI and black box AI. This challenges the notion that attribution of responsibility is decreased by AI transparency and highlights the complex nature of this phenomenon. The study challenges common thinking by showing that the success of a decision outcome does not significantly impact responsibility attribution, inferring that accountability stays relatively constant in managerial decision-making regardless of the outcome. In conclusion, this thesis emphasizes the crucial role of human decision-makers in managerial settings and promotes continuous investment in human ethical decision-making training. These findings provide an important contribution to the discussion of AI ethics and responsibility in decision-making.

Date of Award	16 Oct 2023
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Cristina Mendonça (Supervisor)