Research map / Image generation / Autoregressive and latent generation
The Value of AI Guidance in Human Examination of Synthetically-Generated Faces
Aidan Boyd, Patrick Tinsley, Kevin Bowyer, Adam Czajka
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 |
|---|---|---|---|
| Show humans the AI's classification decision | Human examination of synthetic faces · Synthetic face images | human accuracy: 56% -> 59% | Improved |
| Show the human-guided AI's decision rather than the traditional model's decision | Human examination of synthetic faces · Synthetic face images | human accuracy: 59% -> 61% | Improved |
| Show CAM saliency without the AI decision | Human examination of synthetic faces · Synthetic face images | human accuracy: 56% | No clear effect |
| Add CAM saliency to the AI decision | Human examination of synthetic faces · Synthetic face images | human accuracy: 59% -> 56% | Hurt |
| Train the detector with human salience guidance | Human-AI synthetic face examination · Synthetic face images | human accuracy: 59% -> 61% | Improved |
| Train the detector with human salience guidance | Synthetic face examination using CAM · Synthetic face images, CAM Only | MSE between human annotations and model saliency: 0.24 -> 0.13 | Improved |
| Give no AI assistance | Human examination of synthetic faces · Synthetic face images | accuracy: 51.88%±15.51% | No clear effect |
| Use a 90%-accurate AI decision instead of the traditional 50%-accurate AI decision | Human examination of synthetic faces · Synthetic face images, AI accuracy unreported | human accuracy: 57.54±16.6 -> 63.23±11.41 | Improved |
| Use a 10%-accurate AI decision instead of the traditional 50%-accurate AI decision | Human examination of synthetic faces · Synthetic face images, AI accuracy unreported | human-AI agreement: 57.3±15.54 -> 47.61±11.2 | Hurt |
| Tell users a 90%-accurate AI is 90% accurate rather than provide no accuracy information | Human examination of synthetic faces · Synthetic face images, actual AI accuracy 90% | human accuracy: 63.23±11.41 -> 65.15±11.74 | Improved |
| Tell users a 60%-accurate AI is 90% accurate rather than 60% accurate | Human examination of synthetic faces · Synthetic face images, actual AI accuracy 60% | human-AI agreement: 55.69±12.92 -> 59.59±16.71 | Improved |
| Tell users a 90%-accurate AI is 60% accurate rather than 90% accurate | Human examination of synthetic faces · Synthetic face images, actual AI accuracy 90% | human-AI agreement: 65.38±12.04 -> 62.37±10.63 | Hurt |
Closest papers by meaning
- AI-Face: A Million-Scale Demographically Annotated AI-Generated Face Dataset and Fairness Benchmark (CVPR 2025 · similarity 0.86)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · similarity 0.86)
- Human and AI Perceptual Differences in Image Classification Errors (AAAI 2025 · similarity 0.86)
- CapHuman: Capture Your Moments in Parallel Universes (CVPR 2024 · similarity 0.85)
- AugGen: Synthetic Augmentation using Diffusion Models Can Improve Recognition (NeurIPS 2025 · similarity 0.85)
- MultiHuman-Testbench: Benchmarking Image Generation for Multiple Humans (NeurIPS 2025 · similarity 0.85)
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.
