Initially known by the general public in the past few years for its great potential for falsification and disinformation, the technology behind deepfake is generating increased business interest among content creators and promises to catalyze the development of hyperpersonalized entertainment products. What are the foreseeable impacts on the media and entertainment industry?
If there is one thing that deepfake teaches us, it’s that the expression “seeing is believing” deserves to be nuanced in light of how digital technologies are evolving.
As we point out in our 2019 Trends Report, this technology manages to trick our senses by using deep learning, a form of artificial intelligence that is derived from machine learning and is used to model data with a high level of abstraction by training generative adversarial networks (or GANs). By applying this technology to image synthesis, deepfake makes it possible to superimpose images and videos created from pre-existing sources over other images and videos. It is therefore relatively easy to obtain credible results.
Up to now, deepfake has been used mainly to interchange images and short videos. However, given the speed at which it is developing, it could soon be applied to more ambitious projects.
A brief history of deepfake
As a reminder, the term deepfake—which is a contraction of the expression deep learning and the term fake—is a variation of the handle of a Reddit user who posted fake pornographic videos falsely featuring public figures in December 2017.
The technology was also introduced to the public through a short video that went viral and in which filmmaker Jordan Peel used the face of the 44th president of the United States, Barack Obama, to insult the current president, Donald Trump. Followed a slew of memes and amateur videos of all sorts, some more convincing than others that were shared abundantly on social networks.
Rightfully criticized as a dangerous evolution of fake news, deepfake has prompted certain observers to highlight its potential dangers: diversion and violation of images, propagation of false information and ‘decredibilization’ of audio and video evidence to name but a few. The rise of this technology also facilitated the development of an industry based on the detection of fake videos. However, although it is true that deepfake’s development deserves to be cautiously explored with caution, the fact remains that it has major potential in terms of legitimate commercial exploitation—as have pointed out some of Hollywood’s major special effects studios having expressed interest in this technology.
The democratization of costly processes
Although image synthesis has long been a part of the media and entertainment industry, the revolution announced by deepfake’s apologists regards cost and access.
Until recently, switching faces with convincing results was reserved for the greatest of Hollywood productions. For example, following the “digital resurrection” of dead actor Peter Cushing allowing him to resume his role as Grand Moff Tarkin in the film Rogue One: A Star Wars Story, Industrial Light & Magic studio’s creative director claimed that the effect had been extremely difficult and costly to develop. His team had had to resort to a mixed technique combining computer-assisted movement capture and image synthesis systems. By contrast, last year, a YouTuber known for using deepfake uploaded a short video in which he had replaced the face of actor Alden Ehrenreich with the face of a young Harrison Ford in several excerpts of the film Solo: A Star Wars Story. The YouTuber did it alone and practically at no cost! If there is a difference in the quality of both productions, creating such an effect on such a small scale would have been unthinkable a few years ago.
It is worth pointing out that FakeApp, one of the most popular deepfake applications, is available free of charge. It uses the TensorFlow open-source platform, itself made available to the public free of charge by Google in 2015. Of course, obtaining credible results depends on having access to a sufficient quantity of training data of sufficient quality and engaging minimal IT costs to design and deploy models. Here again, significant improvements are foreseeable in the coming years, as recently demonstrated by the researchers at Samsung’s Moscow-based AI centre who successfully animated faces in a convincing manner from simple portraits. Such an easy access to credible special effects as deepfake allows is therefore not only impressive, but also promising when it comes to improvements in the short term.
New products, new opportunities
Given these technological developments, some observers emphasize that the future of the media and entertainment industry among several other industries will necessarily involve the hyperpersonalization of content and increasingly intense interactions between consumers and contents. According to an Accenture report, 58% of consumers would change half or more of their expenses and choose to do business with vendors proposing personalized experiences without compromising their trust. In the same vein, the report titled Perspectives from the Global Entertainment & Media Outlook 2018–2022, produced by PwC, cites personalization among the five main factors of change in the media and entertainment industry and specifies that it’s a key factor that will directly impact the viability of the industry’s different actors.
Well beyond the simple personalized recommendations that platforms such as Netflix got us used to, recent developments in artificial intelligence will soon allow for the large-scale production of intelligent content presenting visual and sound components, characters as well as complete narrative threads adapted to the behaviours and preferences of individual consumers. Among deepfake’s potential contributions to this trend, one can imagine the possibility for viewers to replace the image of one actor by the image of another actor in a film or to enable video game players to create ultra-sophisticated avatars in their image.
It is no surprise that the porn industry is leading the way when it comes to commercializing hyperpersonalized products incorporating deepfake. An American production company unveiled excerpts in which it demonstrated its ability to switch faces between two actors, to transpose an actor in a different location and even to include a viewer in one scene. A multiplication of similar initiatives elsewhere in the industry seems inevitable sooner or later.
Having recognized the strategic importance of high-quality data for the future of image synthesis, some companies already offer actors and public figures high-priced “digital preservation” services that enable them to recreate their image in future productions and thus artificially preserve their youth.
Given the increased accessibility of credible special effects made possible by deepfake, it would not be surprising to witness the appearance of business models based on the widespread licensing of actors’ or celebrities’ images. Just as with Robin Wright’s character in the film The Congress, certain celebrities could decide to entrust the commercial exploitation of their image to specialized managers and thereby appear in dozens of television or video game productions—big and small—without having directly participated in them.
Some food for thought
Beyond the incalculable number of ethical discussions about the issue of deepfake, its commercial exploitation also raises several legal questions.
Like the other technologies based on artificial intelligence, deepfake raises questions with respect to the rights and recourses that intellectual property rights holders have with respect to the training data used. On a more concrete level, if most of the deepfake models that are designed and deployed by simple users to copy the appearance of celebrities use copyrighted works (including, for example, films or television series) as training data, the legality of this practice is far from certain.
Although the questions regarding the interaction between copyright and artificial intelligence have been raised in several modernization projects undertaken by certain states and organizations, including the European Union which recently adopted a new directive on copyright including a limited number of exceptions regarding text and data searches, the courts have not really had to deal with these issues up to now.
Seeing as deepfake makes it possible to exploit the image of people who have died, its development risks feeding discussions on the right to one’s image after death. Such discussions are already the object of public debate given the growth of the dead celebrity hologram industry. May it, however, be said that there are large grey areas and major legislative differences in that regard between different jurisdictions.
Even in the present era of rapid technological developments, it is rare to come across a technology that is as preoccupying as it is fascinating as deepfake. Deepfake’s ramifications stem from political disinformation to cinema. However, regardless of what the years to come have in store for deepfake in political, legislative and commercial terms, chances are that it will become increasingly common on our screens.