Error as insight: AI hallucinations and the pedagogical possibilities of generative visual misrepresentation

This theoretical paper reframes visual artificial intelligence (AI) ‘hallucinations’ not as errors to be corrected but as productive sites for reimagining how visual meaning is constructed, transmitted, and contested. Drawing from established visual literacy frameworks, the paper proposes that AI’s failures in historical visualization reveal fundamental insights about the nature of visual knowledge production and representation. Building on perceptual, structural, and ideological analytical perspectives, the analysis examines how AI systems trained on mainstream visual datasets produce errors that expose the limitations and biases in dominant systems of visual representation. These technological glitches become pedagogical opportunities for developing critical visual literacy skills in an age of algorithmic image generation.