Beyond human perception: challenges in AI interpretability of orangutan artwork

Area of Science:

• Comparative psychology
• Artificial intelligence
• Evolutionary anthropology

Background:

• Artistic expression is a unique window into the cognitive and emotional states of artists.
• Non-human primates, such as orangutans, create non-figurative drawings that offer potential insights into their minds.

Purpose of the Study:

• To apply artificial intelligence (AI) and interpretability tools to analyze seasonal nuances in an orangutan's drawings.
• To explore the potential of AI in understanding non-human cognitive and emotional expressions through art.

Main Methods:

• Utilized Convolutional Neural Networks (CNNs), specifically VGG19 and ResNet18 models, for drawing analysis.
• Employed the Captum interpretability tool, including occlusion, integrated gradients, PCA, t-SNE, and Louvain clustering.
• Focused on decoding seasonal influences within the non-figurative drawings of an orangutan named Molly.

Main Results:

• Achieved notable accuracy in classifying seasonal patterns within the orangutan's drawings.
• Identified critical areas and elements in the drawings that influence seasonal recognition.
• Demonstrated AI's capability to reveal complex influences beyond traditional human-centric analytical methods.

Conclusions:

• AI analysis of non-human art provides significant insights into non-human cognition and emotional expression.
• This approach advances the study of animal art and has implications for evolutionary anthropology and comparative psychology.
• The findings highlight the potential of AI to deepen our understanding of diverse forms of artistic expression across species.