Artificial intelligence can emulate human normative judgments on emotional visual scenes.

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Affective Computing

Background:

• Human emotion concepts develop through biology, language, and cognition.
• Multimodal artificial intelligence (AI) models show advanced intellectual abilities.
• AI's capacity for affective response to stimuli remains largely unexplored.

Purpose of the Study:

• To investigate if state-of-the-art multimodal AI systems can emulate human emotional ratings.
• To assess AI's ability to judge affective dimensions and discrete emotions in images.
• To explore the emergence of emotion concepts in AI through statistical learning.

Main Methods:

• Utilized state-of-the-art multimodal AI systems.
• Presented a standardized set of images as affective stimuli.
• Compared AI-generated emotional ratings against average human ratings.

Main Results:

• AI judgments showed a strong correlation with average human emotional ratings.
• The AI systems were not explicitly trained to match human affective responses.
• The findings suggest emotion concept development can emerge from large-scale image-text data.

Conclusions:

• Multimodal AI can develop the ability to visually judge emotional content.
• Language and statistical learning are key to fostering emotion concepts in AI.
• These findings have significant implications for the responsible deployment of AI technology.