Reproduce “Improved Emotional Alignment of AI and Humans: Human Ratings of Emotions Expressed by Stable Diffusion v1, DALL-E 2, and DALL-E 3” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
James Derek Lomas, Willem van der Maden, Sohhom Bandyopadhyay, Giovanni Lion, Nirmal Patel, Gyanesh Jain, Yanna Litowsky, Haian Xue and 1 more
Researchers from Delft University of Technology quantitatively assessed the capacity of Stable Diffusion v1, DALL-E 2, and DALL-E 3 to express human emotions as perceived by human raters. They found DALL-E 3 achieved the highest emotional alignment (M=7.04), significantly outperforming earlier models, with expressions on human figures generally perceived as better than those on robots, and identified varying AI proficiency across different emotions.
Key insights
- A clear performance hierarchy exists among generative AI models for emotional alignment, with more advanced models demonstrating improved capabilities.
- AI models exhibit lower emotional alignment when depicting emotions on robotic figures compared to human figures, highlighting a challenge for embodied AI.
- The ability of AI to express emotions varies significantly by the specific emotion, indicating differing 'emotional granularity' and suggesting some emotions are inherently easier for AI to portray accurately.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.