Amid the swift advancement of AI technology, the issue of creative work infringement has taken center stage. A collaborative research team from Stony Brook University (part of the State University of New York system) and Columbia University recently published a groundbreaking paper. Their research supports the hypothesis that "AI, specifically trained using high-quality copyrighted creative works, exhibits a superior understanding of readers' preferences compared to professional writers, all at a substantially lower cost."

During the study, the scientific team chose mainstream AI systems, exemplified by GPT-4o. They generated texts that mirrored the writing styles of 50 acclaimed authors, including Nobel laureates in literature. Simultaneously, they invited professional writers to craft similar pieces. The findings indicated that AI-generated works, which had undergone targeted fine-tuning, were more popular among readers than those created by human professional imitators.

In the evaluation phase, 159 participants (comprising 28 writing experts and 131 non-experts) were tasked with comparing the styles and qualities of the works, unaware of the authors' identities. The results were striking: experts were eight times more likely to select AI-generated texts based on style and twice as likely to do so based on quality. Non-expert groups also demonstrated a marked preference for AI-generated works.

Furthermore, modern AI detection tools boasted a 97% accuracy rate in identifying standard AI outputs. However, they could only pinpoint 3% of the fine-tuned works, highlighting the sophistication of the AI's adaptation.

The study also underscored the cost-effectiveness of AI. Training an AI model to emulate a single writer's style averaged just $81. In contrast, the creation cost for human professionals to produce imitative texts of comparable quality soared to $25,000, marking a staggering 99.7% reduction in expenses when utilizing AI.