澳大利亚维多利亚州选举前出现多个AI生成的深度伪造视频以付费广告形式传播，内容涵盖虚假的暴力事件和灾害场景，包括"加油站砍刀攻击""妇女在路边分娩"和"洪水冲进议会"等场景1。
研究人员对2018至2026年间的19项研究进行了元分析，涉及超过24000名参与者，旨在评估公众识别深度伪造内容的能力1。分析结果显示，虽然约75%的样本能正确识别深度伪造内容，但仍有约四分之一的人无法确定其真伪1。更令人担忧的是，近29%的参与者错误地将真实视频标记为伪造，18%表示不确定1。此外，近50%的参与者表示无法自信准确地识别真实政治内容1。
研究还揭示了识别能力与个人特征的关联1。左倾人士在检测涉及与其政治立场一致的政治人物的伪造内容时可能表现更准确，但这种优势仅在特定情况下出现1。值得注意的是，高度新闻消费者显示出信心与准确性的脱节——他们自认为能够判断政治内容，但实际表现不佳1。这一现象威胁到民众对民主制度的信任。
Deepfake videos have emerged as a significant threat to this year's Victorian election, with artificial intelligence-generated content depicting false violent incidents and disaster scenarios circulating as paid advertisements 1. A meta-analysis of 19 studies involving over 24,000 participants conducted between 2018 and 2026 found alarming gaps in the public's ability to distinguish manipulated from authentic political content 1.
The research revealed that approximately one-quarter of participants could not accurately identify deepfakes, while nearly 29 percent incorrectly flagged genuine videos as fabricated 1. The false scenarios documented included fabricated scenes of gas station knife attacks, women giving birth roadside, and floodwaters breaching parliament 1. Additionally, close to 50 percent of participants expressed low confidence in their ability to accurately judge authentic political material 1.
The study identified a notable partisan dimension to deepfake detection: left-leaning individuals showed marginally improved accuracy when identifying manipulated content portraying political figures aligned with their own political positions 1. However, a troubling pattern emerged among heavy news consumers, who displayed strong confidence in their judgment despite demonstrating poor actual performance in distinguishing real from fabricated political content 1. These findings underscore the threat deepfakes pose to public trust in democratic institutions as voters face increasing difficulty in navigating a landscape where visual evidence can no longer be reliably assumed authentic 1.
评论
还没有评论，欢迎留下第一条。