Abstract
Current implementations of artificial intelligence (AI) ethics do not adequately consider feelings or affect. If AI should align with human ethics, it is essential to explore the possibility of AI behavior mirroring virtuous human ethical conduct, where feelings play a role in actions, judgments, or statements.
Background
While prominent theories of normative ethics often highlight their differences and shortcomings, Virtue, Consequentialist, and Kantian Deontological ethics all share a common feature of considering human feelings to some degree. The popular descriptive ethics theory, Moral Foundations Theory, positions feelings as central to many of its foundations.
Dataset Construction
In this paper, a moral valence dataset is proposed, consisting of 500 annotations by six human participants for both action/judgment and consequence moral valence, ranging from -1 to 1 for text-presented scenarios from the Commonsense Norm Bank dataset. The resulting valence features show significant relationships with multi-class (immoral/discretionary/moral) and binary immoral/moral categories.
Results and Analysis
Using regularized logistic regression for binary classification, a noteworthy Matthews correlation coefficient of 0.764 is achieved, providing early evidence for the usefulness of valence features in morality estimation of text. This indicates that considering the valenced consequences of responses for others can contribute to more human morally-aligned AI.
Future Directions
In the interest of promoting further affective-moral computing research, this study's annotations will be made available for research on request.
Blogger's Review: This paper explores the relationship between emotion and morality, providing a fresh perspective for AI ethics research. By quantifying emotional features, the authors lay the groundwork for developing AI that aligns better with human moral standards, making future investigations highly anticipated.