The training gain was driven by a 14.2-percentage-point improvement in accuracy on real images, while the gain on AI-generated images was not statistically significant¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? Accuracy on real images increased by 14.2 percentage points (95% CI: [0.7, 27.7]) after training, a corresponding reduction in false-positive errors on real images. → Negar Kamali 2026 q3 i? Accuracy on AI-generated images showed a positive but non-significant estimated increase of 4 percentage points (95% CI: [-5.8, 13.8]). → Negar Kamali 2026 (2)

Evidence¶

Negar Kamali 2026¶

Negar Kamali, Candice Rockell Gerstner, Jessica Hullman, Matthew Groh. (2026). Generative AI Literacy Training Improves Intelligence Analysts' Discrimination of Real and AI-Generated Images. https://arxiv.org/abs/2606.28510 doi:10.48550/arxiv.2606.28510

q3 · i? · causal · r2

Pre-registered OLS estimate for the real-image subset of the counterbalanced experiment (Table 1, column 2). The article reports accuracy on real images "increased by 14.2 percentage points (95% CI: [0.7, 27.7])", indicating sharper discrimination rather than a conservative shift.

"Instead, accuracy on real images increased by 14.2 percentage points (95% CI: [0.7, 27.7]), corresponding to a 14.2 percentage point reduction in false-positive errors and indicating sharper discrimination capabilities."

Negar Kamali 2026 (2)¶

Negar Kamali, Candice Rockell Gerstner, Jessica Hullman, Matthew Groh. (2026). Generative AI Literacy Training Improves Intelligence Analysts' Discrimination of Real and AI-Generated Images. https://arxiv.org/abs/2606.28510

q3 · i? · causal · r2

Logistic-regression robustness check on the AI-generated-image subset. The article reports the effect "is positive but not significant (β= 0.331,p= 0.330)"; the OLS estimate was a 4-percentage-point increase with CI [-5.8, 13.8] per the Discussion. No statistically significant gain was detected on AI-generated images.

"the training effect is larger and statistically significant on real images (β= 0.631,p= 0.040), whereas the effect on AI-generated images is positive but not significant (β= 0.331,p= 0.330), consistent with the interpretation that training sharpened discrimination without inducing a conservative bias toward labeling images as fake."

Discussion¶

Related Claims¶

• A 30-minute generative AI literacy training increased intelligence analysts' overall accuracy at distinguishing real from AI-generated images by 9 percentage points — related
• Training effects concentrated in portrait images (+39.8 percentage points, significant), with directional non-significant negative estimates for full-body and posed-group images — related
• Training gains were distributed broadly across the 97 stimulus images rather than concentrated in a small subset, though stimulus-level effects were heterogeneous — related
• Training improved discrimination sensitivity (d′) with only a small, non-significant criterion shift toward labeling images as AI-generated — related
• Training improved accuracy in every digital-forensics experience group and most generative-AI familiarity groups, with the largest estimated gains among analysts with no forensics experience and frequent generative-AI users — related