ABSTRACT Deepfakes are a form of human image synthesis where an existing picture or image is superimposed into a video to change the identity of those depicted in the video. The technology relies on machine learning or artificial intelligence to map an existing image, usually a photo of a person's face, to transfer that image to an existing video image. The technology emerged into the latter part of 2017, and has since given rise to apps and other programmes that allow users to create their own deepfakes. We already use filters and emojis to alter images by consent, however, deepfakes are particularly problematic because they allow for production of videos that are highly convincing, taken to be a real video of the person depicted. Deepfakes provide for the manipulation of all manner of video, but particular risks include videos produced to incite political deception, voter manipulation, commercial fraud, and ‘revenge porn’. The production of deepfake ‘revenge porn’ presents as especially insidious given the ability to transfer the face of any person onto an already existing pornographic video. Harm is exacerbated where that video is then disseminated, via the internet or by social media.

中文翻译：

---

Deepfakes 和图像操纵：刑事定罪和控制

摘要 Deepfakes 是人类图像合成的一种形式，其中将现有图片或图像叠加到视频中以改变视频中描绘的人的身份。该技术依靠机器学习或人工智能来映射现有图像（通常是人脸照片），以将该图像转换为现有视频图像。该技术于 2017 年下半年出现，此后催生了允许用户创建自己的深度伪造的应用程序和其他程序。我们已经使用过滤器和表情符号通过同意来改变图像，但是，深度伪造尤其成问题，因为它们允许制作非常有说服力的视频，被视为所描绘人物的真实视频。Deepfakes 提供了对各种视频的操作，但特定风险包括为煽动政治欺骗、操纵选民、商业欺诈和“报复色情”而制作的视频。鉴于能够将任何人的面部转移到已经存在的色情视频上，deepfake“复仇色情”的制作显得尤为阴险。在该视频随后通过互联网或社交媒体传播时，危害会加剧。