EmoLab
  • Home
  • People
  • Research
  • Publications
  • Resources
  • Join Us

Human × AI Affective Image Space

← Back to EMOLAB Research
EMOLAB Emotion Space Atlas

Human × AI Affective Image Space

How humans and ChatGPT place the same MATTER images in valence–arousal space

Kim (2026) · Perception View paper ↗
Explore

View controls

View

Category means

Pair lines

Pair lines appear only in Overlay view and connect the same image across Human and ChatGPT ratings.

Highlight

Affective image space

Humans

224 images
Valence
Arousal
Shape = source Color = image category Dashed line = same image across sources
Guide

Read the space

Selected image

Hover over a point to compare Human and ChatGPT ratings for that image.

Category mean

Select a category to compare its mean position across sources.

Legend

Human
ChatGPT

Axes

Valence1–9

Negative ↔ Positive

Arousal1–9

Low ↔ High

Interpretation

How the spaces differ

01

Lower arousal

ChatGPT placed all four category means lower on arousal, with the largest drops for fearful and disgusting images.

02

Mirthful shift

Mirthful images moved strongly toward lower valence in ChatGPT ratings (ΔV = −1.81).

03

Disgusting shift

Disgusting images moved toward higher valence in ChatGPT ratings (ΔV = +1.08), while arousal decreased.

224 images 4 categories 2 sources 1–9 rating scale Valence × Arousal space

EmoLab · Department of Psychology · Jeonbuk National University

 

Jeonju, Republic of Korea · Site Editor