Generative AI Has a Visual Plagiarism Problem

Large language models (LLMs) and image generation AI systems can produce outputs that closely resemble or directly copy copyrighted material, even without explicit prompts to do so. Experiments with Midjourney and DALL-E 3 revealed they can generate near-identical images of trademarked characters and film scenes. This raises serious concerns about copyright infringement, as users may unknowingly create infringing content. Companies like OpenAI and Midjourney likely trained their systems on copyrighted materials without proper licensing or artist consent. Potential solutions include retraining models without copyrighted data or implementing better filtering, but these present technical challenges.