Research map / Large language models / AI and society
HIVE: Evaluating the Human Interpretability of Visual Explanations
Sunnie S. Y. Kim, Nicole Meister, Vikram V. Ramaswamy, Ruth Fong, Olga Russakovsky
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| provide explanations to participants (agreement task) | GradCAM, BagNet, ProtoPNet, ProtoTree · CUB agreement | belief that incorrect predictions are correct: 67.2%, 57.6%, 53.6%, 62.8% of incorrect predictions believed correct | Hurt |
| provide explanations on ImageNet (agreement task) | GradCAM, BagNet · ImageNet agreement | belief that incorrect predictions are correct: 55.2% and 64.4% | Hurt |
| switch from correct to incorrect predictions in distinction task | GradCAM · CUB distinction | participant accuracy (%): 71.2 -> 26.4 | Hurt |
| switch from correct to incorrect predictions in distinction task | BagNet · CUB distinction | participant accuracy (%): 45.6 -> 32.0 | Hurt |
| show ground-truth class labels (ablation) | GradCAM and BagNet · ImageNet distinction | participant accuracy (%): 51.2 -> 49.2 (GradCAM), 38.4 -> 34.8 (BagNet) | Hurt |
| compare model similarity scores with human similarity ratings (ProtoTree) | ProtoTree · CUB | Pearson correlation: 0.06 | No clear effect |
| compare model similarity scores with human ratings (ProtoPNet) | ProtoPNet · CUB | Spearman correlation: rho=-0.25, p=0.49 (distinction); rho=-0.52, p=0.12 (agreement) | No clear effect |
| replace similarity-rating agreement task with selecting first disagreeing tree step (ablation) | ProtoTree · CUB agreement | participant accuracy (%): 53.6 -> 52.8 | No clear effect |
| use automatic localization metrics to predict human agreement confidence | GradCAM · CUB/ImageNet agreement | correlation with human confidence | No clear effect |
| show participants their task performance before final self-rating | self-rated understanding · all studies | self-rated understanding (1-5): 3.8 -> 3.5 | Hurt |
| increase deployment risk level in tradeoff question | baseline vs explained model · tradeoff questions | required extra baseline accuracy: +6.2% -> +8.2% -> +10.9% | Improved |
| compare ProtoTree on distinction against others | ProtoTree · CUB distinction (correct) | participant accuracy (%): 71.2 (GradCAM) vs 33.8 | Hurt |
Closest papers by meaning
- HumanPCR: Probing MLLM Capabilities in Diverse Human-Centric Scenes (ICLR 2026 · similarity 0.88)
- I-CEE: Tailoring Explanations of Image Classification Models to User Expertise (AAAI 2024 · similarity 0.87)
- Counterfactual Explanations for Misclassified Images: How Human and Machine Explanations Differ (Abstract Reprint) (AAAI 2024 · similarity 0.87)
- Navigating the Maze of Explainable AI: A Systematic Approach to Evaluating Methods and Metrics (NeurIPS 2024 · similarity 0.87)
- Position: Explainable AI Cannot Advance Without Better User Studies (ICML 2025 · similarity 0.86)
- A Multimodal Automated Interpretability Agent (ICML 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
