Agentic data science with the Data Agent Kit
cloud.google.com·8h ago
AI Research
arXiv CS.AI·Jul 24·5 min read
Can Valence Reflect Morality in Natural Language? A Preliminary Annotation Study
TL;DR
A dataset of moral valence was created, consisting of 500 annotations to assess how feelings influence moral judgments in AI. The findings suggest that valence features can effectively estimate morality in text, with a notable correlation coefficient of 0.764 for binary classification.
✦ Why It Matters
Consider integrating moral valence features into your AI models to enhance ethical alignment.
Key Takeaways
- Explore the role of feelings in AI ethics frameworks.
- Utilize the moral valence dataset for your research.
- Implement regularized logistic regression for classification tasks.
- Investigate the relationship between valence and moral categories.
Full Summary
Current AI ethics frameworks often overlook the role of human feelings in moral decision-making. This study introduces a dataset that captures moral valence, which ranges from -1 to 1, based on 500 annotations from six human participants evaluating various scenarios.
The research utilized text from the Commonsense Norm Bank and focused on both action/judgment and consequence moral valence. Results indicated significant relationships between valence features and moral categories, achieving a Matthew's correlation coefficient of 0.764 using regularized logistic regression for binary classification.
These findings provide early evidence that incorporating valence can enhance AI's alignment with human moral standards. The dataset will be made available for further research in affective-moral computing.