讀者偏好 ChatGPT 故事勝過人類作者，但在得知 AI 標籤後態度反轉
一項針對 2500 名參與者的研究發現，讀者在未知來源時給 ChatGPT 短篇小說的品質評分高於人類作品，但一旦得知 AI 身份評分便會驟降，揭示了人們對機器創作的明顯偏見。
a study published in the journal judgment and decision making analyzed the reactions of 2,587 participants who read six short stories, half written by humans and half generated by chatgpt-4. when the source was blinded, ai stories received significantly higher mean scores for quality (1.54 vs 0.97) and immersion (1.42 vs 1.00) than human texts, likely because ai writing is smoother and easier to process. however, the experiments revealed a robust 'ai labeling effect': when participants were told a story was ai-generated, their ratings dropped, and when told it was human-written, ratings rose, regardless of the actual origin. furthermore, ai skeptics rated stories lower when told they were ai-generated, while ai enthusiasts rated them higher, yet no group could reliably identify the stories' origins better than chance when tested directly.