How to Film Bruce Willis in His Absence: Deepfake Creation and Application

Photo: Megafon / YouTube

In recent years, advanced technologies for creating deepfake images have made it almost impossible to distinguish them from real photos and videos. They can help to recreate stars of the past, to insert images of famous people into places they’ve never been, and to convincingly use audio deepfakes instead of real calls. Researchers discussed the future development of deepfakes and how to protect yourself from new types of fraud during the round-table discussion ‘Fake News: An Interdisciplinary Approach’ as part of the XXIII Yasin (April) International Academic Conference on Economic and Social Development.

Deepfake uses synthetic media based on artificial intelligence and the way it works is based on the idea that one artificial intelligence can deceive another, says Alexey Neznanov, Associate Professor at HSE School of Data Analysis and Artificial Intelligence. For example, if you add a sticker or a special filter to a picture of a panda, then a neural network trained for animal recognition won’t see the panda, but might see a gibbon instead.

Alexey Neznanov, Associate Professor at HSE School of Data Analysis and Artificial Intelligence

‘There are Generative Adversarial Network (GAN) services that are learning to reproduce an object more accurately so that the recognizer identifies it more accurately. Anyone can work on developing these, you only need a basic knowledge of mathematics (1^st-year undergraduate level) and Internet access.’

The principles of deepfake are based on the technology of object recognition and its replacement. In 2019, the first-generation of relatively convincing deepfakes of politicians and Hollywood stars appeared.

By 2021, deepfakes were being easily produced and commonly used in cinema and advertising. The most famous case in Russia concerned an advertisement for Megafon, when Hollywood actor Bruce Willis literally "traded his face” with his image being inserted into the video, without him ever acting in person in front of the camera, via a neural network.

Now we are on at least the third generation of deepfakes. With each new stage of technological development, the rendering of faces, realism, and functionality improves. Alexey Neznanov believes that the 4th generation of deepfakes is coming soon with will tend towards an omnichannel focus (video, text, smells and other identifying characteristics at the same time).

There are many ways to use deepfakes for good — from advertising to the transformation of museum experiences, says the expert.

There are examples of legally successful deepfake startups. For example, the Russian company Deepcake, in addition to the Megafon ad with Bruce Willis, recreated the face of George Miloslavsky, a character from the movie ‘Ivan Vasilievich: Back to the Future’ as portrayed by Leonid Kuravlev, for a SBER advert. For this project, the training of artificial intelligence lasted 16 days, and the programme processed about 7000 images.

However, it would be strange if this technology of replacing faces in photos and videos was used only legitimately and for good. ‘A sharpened metal blade can be a life-saving scalpel in the hands of a surgeon, or it can be a weapon. The same drug can be both a poison and a medicine. There are hundreds of examples of this kind, and it can also be applied to deepfakes. It’s all about the application of the technology,’ says Dmitry Ogorodov, lawyer, and expert of the Committee on Artificial Intelligence at the Commission of the Russian Federation for UNESCO.

The technology opens up a lot of opportunities for a wide range of nefarious activities.

Dmitry Ogorodov highlighted the main ones:

• audio deepfakes for the purpose of extortion (for example, calls like ‘Mum, I’m in trouble, please send money to this number’, ‘your account is blocked, provide your PIN code’, etc.);
• perjury and falsifying evidence in civil and criminal proceedings (for example, someone creates a video showing a person declaring that they borrowed money and are not going to give it back, however there is a receipt stating that they fully undertake to pay the debt);
• defamation of character (for example, a fake image of a politician visiting a nightclub, where they should not be).

Provision for the responsibility for such violations and falsifications has already been stipulated in the legislation of the Russian Federation. The problem is the capability of existing forensic examination. ‘Many methods used by experts in ordinary cases become ineffective with the development of deepfake technologies,’ notes the researcher. ‘In such cases, forensic experts have to admit they can't determine whether it's true or not,’ he states.

‘All my colleagues state: ‘We are not ready.’ Electronic signatures and watermarks can go some way towards solving the problem, but they are still not widely used, says Alexey Neznanov.

He sees the ‘Detect Fakes’ project as one of the solutions to the problem. On this website, ordinary users help to create a database of deepfakes and real pictures for further fact-checking. This initiative is open to everybody.

More radical measures remain controversial, say the experts. During the round table, the researchers proposed limiting the circulation of photos and videos which neural networks can learn from, or limiting the availability of processors and video cards. There were also ideas about creating an international organization to control the spread of deepfakes, similar to the IAEA. This approach shows that fake videos in the modern era are becoming just as dangerous as weapons-grade plutonium.