When AI fabricates science: trust hinges on image provenance
AI-made scientific images can look convincingly real, challenging journals and the public to tell them apart and risking a broader crisis of trust in science. High-profile cases—AI-generated figures in 2024 papers and an AI-modified image triggering a 2026 NEJM retraction—show how detectors can lag behind image creation. As visual credibility has long rested on provenance, institutional authority, and alignment with observed data, generative AI erodes those cues. The path forward is transparency: clear disclosure of image provenance (AI-generated or not), explicit explanations of what the image represents, verification and reproducibility details, and cross-field standards for image integrity. Ultimately, public trust depends on documenting the link between visuals and verifiable scientific reality, not on sleek visuals alone.
- Anyone can fake a scientific image with AI, tricking even academic journals – and undermining trust in science The Conversation
- AI-generated images undermine scientific visual evidence Let's Data Science
- The Real Threat Isn't Deepfakes—It's the Collapse of Trust": Amitabh Kumar on Building Contrails AI Indian Startup Times
- New chip fights deepfakes with built-in cryptographic signing Interesting Engineering
Reading Insights
1
11
12 min
vs 13 min read
95%
2,427 → 116 words
Want the full story? Read the original article
Read on The Conversation