Unveiling The Factors of Aesthetic Preferences with Explainable AI

Machine learning can identify which visual attributes matter most for aesthetic preferences: The study used multiple ML models (Random Forest, XGBoost, and others) to predict how aesthetically pleasing images are based on specific visual attributes, demonstrating that computational methods can quantify factors humans find beautiful.

SHAP explainability reveals why models make aesthetic judgments: By applying the SHAP technique, researchers could interpret which attributes most influenced each prediction, moving beyond "black box" decisions to show stakeholders the specific visual features driving aesthetic scores—important for building trust in AI systems that evaluate creative work.

Different attributes interact in complex ways to shape aesthetic preferences: The research examined three major image databases (AADB, EVA, and PARA) and found that aesthetic judgments aren't determined by single factors but by combinations of attributes working together, highlighting the nuanced nature of human aesthetic preferences.

Explainable AI can democratize understanding of aesthetic decision-making: By making the factors behind aesthetic judgments transparent and interpretable, this approach allows non-technical stakeholders—such as photographers, designers, or content creators—to understand what computational systems consider aesthetically valuable, rather than treating these judgments as mysterious algorithmic outputs.

Responsible AI requires both accuracy and interpretability: The study's emphasis on comparing model performance alongside SHAP explanations demonstrates that for applications affecting creative decisions, simply achieving high prediction accuracy is insufficient—systems must also explain their reasoning to be trustworthy and useful.