 |
| January 2026 |
 |
75 |
|
| 研究論文Research Articles |
| 眼見不一定為實：生成式人工智慧深度偽造圖片的真實性認知研究 |
| Seeing Is Not Always Believing: A Cognitive Study on the Authenticity of Deeply Falsified Images by Generative Artificial Intelligence |
|
|
|
 |
(33)
|
| 作者 |
白龍、董飛飛、馮迪拉 |
| Author |
Long BAI, Feifei DONG, Dira FENG |
| 關鍵詞 |
生成式人工智慧、深度偽造資訊、智能欺騙、虛假資訊 |
| Keywords |
Generative artificial intelligence, deep-fake information, intelligent deception, false information |
| 摘要 |
生成式人工智慧(artificial intelligence, AI)深度偽造技術創造出的智能欺騙，衝擊人類「有圖有真相」、「目知眼見」的真實性認知傳統。研究考察229位受訪者對深度偽造圖片的真實認知表現，發現識別AI深度偽造圖片的準確率僅有39.3%。性別類型和AI技能經歷的不同，並不會導致甄別水準有顯著差異，但受訪者觀看文化接近性不同的視覺內容在辨識水準上有顯著差異。研究認為人工智慧生成內容（artificial intelligence generated content, AIGC）深度偽造技術已經超越了恐怖谷理論（uncanny valley）的閾值，虛假線索資訊丟失導致受訪者陷入決策的隨機性，對合成圖像真實性認知的不確定性增強。如未來不能正視真實性危機，人類社會的真實認知將可能被顛覆，「資訊災變」和「現實冷漠」的社會風險就有可能到來。 |
| Abstract |
Intelligent deception created by deep-fake technology in generative AI impacts the human tradition of “pictures do not lie” and “seeing is believing” perception of authenticity. The study evaluates the ability of 229 participants to discern the authenticity of deeply faked images and reveals that accuracy in recognizing AI-generated deepfakes stood at only 39.3%. Differences in gender and experience with AI skills did not lead to significant differences in screening levels, but respondents viewing visual content with varied cultural proximity differed significantly in recognition levels. The study concludes that generative AI’s deep forgery technology has exceeded the threshold of the “uncanny valley,” where the loss of information about false cues leads to randomness in decision-making and increased uncertainty about the perceived authenticity of synthesized images. If this authenticity crisis is not addressed in the future, the perception of truth in human society may be subverted, and the social risks of “information catastrophe” and “reality indifference” may become actualized.
本文引用格式：
白龍、董飛飛、馮迪拉（2026）。〈眼見不一定為實：生成式人工智慧深度偽造圖片的真實性認知研究〉。《傳播與社會學刊》，第75 期，頁119–143。
Citation of This Article:
Bai, L., Dong, F., & Feng, D. (2026). Seeing is not always believing: A cognitive study on the authenticity of deeply falsified images by generative artificial intelligence. Communication and Society, 75, 119–143. |
|
|
 |
| No.74 2025 October |
 |
| No.73 2025 July |
 |
| No.72 2025 April |
 |
| No.71 2025 January |
 |
| No.70 2024 October |
|
