Kamali, Gerstner, Hullman & Groh (2026) — cs.HC / cs.AI / cs.CY 📄 Full text (arXiv)
Kamali et al. (2026) evaluate a Generative AI Literacy training intervention designed to improve intelligence analysts' ability to distinguish real photographs from AI-generated images. In a controlled experiment, trained analysts significantly outperformed untrained controls on image discrimination tasks, with gains persisting on challenging edge cases. This is an important contribution to ai-literacy research because it demonstrates that detection skills are teachable even among domain experts, challenging the assumption that AI-generated content is fundamentally undetectable. The training draws attention to specific visual artifacts and generative model signatures, providing a template for broader professional-training curricula. The study connects to equity concerns around AI-generated misinformation, since disparities in detection ability can compound existing information-access inequalities. It also informs generative-ai regulation discussions by providing evidence that literacy interventions are a viable complement to technical watermarking approaches. The work extends ai-literacy-continuum-higher-education by showing that literacy training is relevant not just for students but for professionals across domains.