WIPO's virtual exhibition on Artificial Intelligence and Intellectual Property
Focus: the advent of deepfake technology and related legal issues
25 November 2020
On 18 September 2020 the World Intellectual Property Organization (WIPO) launched "AI and IP, a virtual experience", the first online and interactive exhibition on the relationship between IP and AI.
On 18 September 2020 the World Intellectual Property Organization (WIPO) launched "AI and IP, a virtual experience", the first online and interactive exhibition aimed at fostering a more comprehensive understanding of the relationship between IP and AI and develop practical approach policymakers could implement. The exhibition offers visitors the opportunity to explore some of the many ways AI is shaping art, music, entertainment and technology and focuses on the following crucial question: do AI-generated works and inventions fit into the current IP systems?
Section "AI and Entertainment" of the exhibition is dedicated to the well-known phenomenon of deepfake technology, which is at the center of ongoing heated debates involving IP, privacy and security issues and on which this paper focuses its attention.
What does "deepfake" mean and how is it created?
A "deepfake" is human image synthesis techniques based on AI algorithm that generates realistic face-swaps. The AI algorithm captures common characteristics, such as shape, colors and style, from real images (so-called "source image" or "source actor") to endue other images with the captured characteristics. Deepfakes are created through generative adversarial networks (GANs), that use two competing neural networks in order to generate data that can be almost indistinguishable from real data.
The most common form of deepfake involves the generation and manipulation of human imagery.
The exhibition offers an example of AI technology used to create deepfakes: "HeadOn", a reenactment system for portrait videos recorded with a commodity RGB-D camera launched in 2018 by research scientists. According to their report issued on 29 May 2018, "reenactment approaches aim to transfer the motion of a «source actor» to an image or video of a «target actor». Very recently, facial reenactment methods have been successfully employed to achieve highly-realistic manipulations of facial expressions based on commodity video data". Rather than animating a virtual, stylised avatar (as done in video games), these algorithms replace the face region of a person with a synthetic re-rendering, or modify the target image/video under guidance of a 3D face model. This enables changing the expression of a target person and creating a manipulated output video that suggests different content (for example, a person who is sitting still could appear as if he/she is talking).
IP and privacy concerns
Following the video explaining HeadOn's technology, the exhibition continues with a voice asking the following questions:
- should the current copyright system include the protection of deepfakes? If not, is copyright the appropriate instrument to regulate deepfakes?
- To whom should the copyright belong?
- To the "source actor", who inspires the fake?
- To the "target actor", who should have similar hairstyle, face shape and camera angles compared with the source?
- To the author of the deepfake?
Answering to those questions (and to many others) requires to make a preliminary distinction between legitimate and illegitimate use of deepfakes. Such distinction could also help policymakers in assessing the adequate legal protection of people involved in the deepfake process: in particular, the author of deepfake materials and the "source actor".
The underlying technology has been used for many legitimate creative purposes: movie's or TV program's producers used similar technology for some time. More recently, major digital effects studios have used AI to reproduce a famous actor's likeness and overlap this figure on the real performer: this added value to their stories, but it also cut down production time and costs. For example, "Guardians of the Galaxy vol. 2" (2018) showcased a de-aged, 1980s version of Kurt Russell, while in "Rogue One" (2016) AI technology was used to re-create Peter Cushing's character of "Grand Moff Tarkin" from 1977's Star Wars, despite the fact that Mr Cushing died in 1994.
However, it is worth noting that, unfortunately, the most popular use of deepfakes is the creation of manipulated videos for malicious purposes. In fact, advanced machine learning technology was used in 2017 to create face-swapped pornography, pasting real actors' and pop stars' faces over existing performers in explicit movies.
In light of the above, policymakers should not indistinctly ban this technology, but target cases whereby it pursues malicious purposes and structure a legal framework to protect people whose rights are infringed by deepfakes. In Italy, and in Europe as well, we do not have specific legislation governing this phenomenon. However, some legal and technical experts have recommended adapting current laws covering libel, defamation, identity fraud, copyright and data privacy to this phenomenon.
Focusing on IP and data laws, on one side, the author of the invention could invoke copyright protection. On the other side, AI-based technology could be used to infringe publicity and personality rights. In such cases, the injured party could in theory invoke the right of publicity (i.e. the individual's right to control the commercial use of his/her identity, such as name, image, likeness, or other unequivocal identifiers). At the same time, the unlawful usage of third-party personal data (such as images) could rise data privacy concerns, as it happened in the recent investigations commenced by the Italian data protection authority involving a famous messaging application.
Deepfakes could potentially be extremely useful and dangerous at the same time depending on the purpose of their use. Therefore, managing this phenomenon requires a cooperation between legislative bodies, legal and technical experts to closely monitor and study the underlying technology due to ensure adequate protection of the relevant subjects involved.
Key Take-Away Points
- The first virtual WIPO's exhibition on AI and IP offered an overview of benefits and issues entailed by AI technology.
- The impact of deepfakes on IP and data protection matters.
- Highlighed the lack of legislative framework of AI-based technology
- WIPO exhibition: https://wipo360.azurewebsites.net/
- Report issued by Justus THIES (Technical University of Munich), Michael ZOLLHÖFER (Stanford University), Christian THEOBALT (Max-Planck-Institute for Informatics), Marc STAMMINGER (University of Erlangen-Nuremberg) and Matthias NIESSNER (Technical University of Munich) on "HeadOn": http://184.108.40.206/pdf/1805.11729
- Jackie Snow, Deepfakes for good: Why researchers are using AI to fake health data, FAST COMPANY (Jul. 22, 2020, 11:44 PM), https://www.fastcompany.com/90240746/deepfakes-for-good-why-researchers-are-using-ai-for-synthetic-health-data.
- Robin Pomeroy, This iconic filmstar will star in a new movie - from beyond the grave, WORLD ECONOMIC FORUM (Jul. 22, 2020, 11:47 PM), https://www.weforum.org/agenda/2019/11/james-dean-cgi-deepfakes/.