The heuristics gap in AI ethics: Impact on green AI policies and beyond

Abstract

This article analyses the negative impact of heuristic biases on the main goals of AI ethics. These biases are found to hinder the identification of ethical issues in AI, the development of related ethical policies, and their application. This pervasive impact has been mostly neglected, giving rise to what is called here the heuristics gap in AI ethics. This heuristics gap is illustrated using the AI carbon footprint problem as an exemplary case. Psychological work on biases hampering climate warming mitigation actions is specialized to this problem, and novel extensions are proposed by considering heuristic mentalization strategies that one uses to design and interact with AI systems. To mitigate the effects of this heuristics gap, interventions on the design of ethical policies and suitable incentives for AI stakeholders are suggested. Finally, a checklist of questions helping one to investigate systematically this heuristics gap throughout the AI ethics pipeline is provided.