NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·1h ago
AI Research
arXiv CS.AI·14d ago·5 min read
Position: AI/ML Deepfake Research is Misaligned with AI-Generated Non-Consensual Intimate Imagery (AIG-NCII)
TL;DR
AI-generated non-consensual intimate imagery (AIG-NCII) is largely overlooked in deepfake research, which primarily focuses on authenticity. This misalignment neglects the dignity harms faced by subjects, necessitating a shift in research priorities.
✦ Why It Matters
Prioritize research on AIG-NCII to better protect individuals from dignity harms associated with deepfakes.
Key Takeaways
- Shift focus from authenticity detection to addressing dignity harms.
- Update threat models to include subject-centric harms.
- Implement safety guardrails for researchers and subjects.
- Collaborate with experts in sexual violence prevention.
Full Summary
Current research on deepfakes emphasizes epistemic harms, such as issues of truth and authenticity, while largely ignoring the significant problem of AI-generated non-consensual intimate imagery (AIG-NCII). A landscape analysis of influential studies reveals that most technical interventions focus on detecting authenticity rather than addressing the dignity harms experienced by individuals depicted in these images.
The paper argues that simply knowing an image is synthetic does not alleviate the psychological and social harms inflicted on subjects, and may even worsen their situation. Recommendations include updating threat models to incorporate subject-centric harms and integrating AIG-NCII into AI safety research.
The authors stress the importance of implementing safety measures for both researchers and subjects, and suggest collaboration with experts in sexual violence prevention to navigate this sensitive area effectively.