Modelling societal preferences for automated vehicle behaviour with ethical goal functions

Area of Science:

• AI Ethics
• Human-Computer Interaction
• Moral Philosophy

Background:

• Automated vehicles (AVs) require ethical decision-making frameworks.
• Existing frameworks are largely conceptual and lack empirical grounding.
• Societal values must guide AV moral choices.

Purpose of the Study:

• To develop a quantitative Ethical Goal Function (EGF) for AV decision-making.
• To operationalize ethical frameworks using empirical data.
• To integrate the EGF into a Socio-Technological Feedback (SOTEF) Loop for continuous societal alignment.

Main Methods:

• Discrete Choice Experiments (DCEs) with 89 Dutch university students.
• Evaluation of AV moral scenarios based on six ethically salient attributes.
• Multinomial Logit (MNL) model to derive attribute weights representing societal moral preferences.

Main Results:

• The MNL model demonstrated stable attribute weights and 63.8% predictive accuracy.
• The developed EGF effectively translates societal moral judgments into computational parameters.
• The study provides the first empirical operationalization of AV ethical frameworks.

Conclusions:

• The EGF offers a data-driven approach to encoding societal values in AVs.
• The SOTEF Loop facilitates ongoing refinement of AV ethical decision-making.
• Future research should focus on cultural generalizability and diverse presentation modalities.