Deep Fake Technology has rapidly emerged in recent years due to the rapid development of artificial intelligence. Its vast application potential has brought new possibilities for developing art, social, medical, and other fields. Still, it also provides new tools for criminals to spread false videos, intensify social contradictions, incite violence and terrorist actions, and bring many risks to economic security, social security, and other security fields. This article focuses on deep-fake Technology and conducts in-depth research on its primary connotation, the risks and challenges it brings, and the governance measures of significant countries.
What is Deep Fake Technology?
Deepfake Technology is a technical means of forging content such as pictures, audio, and video using artificial intelligence, machine learning, neural networks, and other methods. It is a derivative technology in the development of artificial intelligence technology. Its core principle is to modify information content such as voice, images, and text using algorithms such as generative adversarial networks or convolutional neural networks.
Deep-fake Technology mainly includes face deep-fake technology and voice deep-fake technology. Among them, face-deep fake Technology can be summarized as the following steps:
Steps of Deep-fake Technology
- Crop and preprocess the face in the target image;
- Extract the identity and expression information in the face and generate a fake face through a generative model;
- Render the generated face to the face position in the target image and reconstruct the image.
For voice deepfake technology, audio and text input are required to specify the content and timbre of the target voice. The main steps of voice-deep fake Technology are:
-
- If the method accepts text input, encode it;
- Extract the Mel-Frequency Cepstral coefficients (Mel-Frequency Cepstral Coefficients, M.F.C.C.s);
- Input the preprocessed data into the generative model to obtain the frame-level speech features of the target speech;
- Obtain the target speech through a vocoder and other methods.
Generative Model of Deep Fake Technology
Generative models based on deep learning are the technical basis of deep-fakes. The main generative models are Generative Adversarial Networks (GAN) and Variational Auto-Encoders (V.A.E.)
An adversarial network is a widely used generative model, including a generator and classifier. The generator can generate images by sampling from Gaussian distribution. The training target generates images the classifier cannot distinguish between true and false. The classifier learns whether the generator generates the image.
The generator and the classifier parameters are fixed alternately during the training process. The more commonly used generative adversarial network models include InfoGAN, CycleGAN, etA variational autoencoder, and generative models based on the autoencoder structure. The auto-encoder generally consists of an encoder module and a decoder module. In the deep forgery process, the identity or action is tempered by adding a decoder and an encoder or injecting information into the encoding. The variation autoencoder has variationalise resistance than the autoencoder.
Deep Forgery Detection Technology
The in-depth development of deep forgery technology has seriously threatened social security, network security, and even political security. People have gradually begun to pay attention to countermeasures to cope with the adverse effects of deep forgery technology.
Images and videos’ deep forgery detection technology can be divided into detection methods for forgery traces and data-driven detection methods. For forgery evidence collection, we can directly detect images, primarily image forgery traces such as “image processing forensics, biometrics, fusion traces, temporal coherence, and model fingerprints.”
In addition to directly collecting image forensics on the detected image, some researchers detect forgery traces generated by the fusion step of generating model output results and peripheral background in the deep forgery step. The data-driven detection method uses convolutional neural networks to detect standard face forgery methods (such as face2face) to obtain relevant feature vectors. Then, it uses neural networks to determine whether the input image has been forged based on the feature vectors.
Voice deep forgery detection is generally divided into the front and back end, respectively extracting a. The front end extracts audio using classicized classification m, mixture models, and neural networks to classify the target audio according to acoustic features to determine whether the video’s voice has been forged.
Risks and Challenges of Deep Fake Technology
In 2020, researchers from University College London released a ranking of what experts consider the most severe A.I. crime threats, with deepfakes topping the list. It is necessary to fully grasp and understand the threats posed by this Technology to prevent and promptly respond to the security risks it brings.
Technology Affects International Relations
Deep fakes will undermine the strategic mutual trust of the international community and impact the fragile peace environment. From the perspective of international stability, deep-fakes combined with other disruptive digital technologies will significantly reduce the trust between global and regional multilateral institutions and international relations actors. According to a test by an authoritative American data science company, through a speech generation algorithm, it only takes two hours of corpus and five days of training to simulate a fake extraordinary voice—Trump’s voice declaring war on Russia. State and non-state actors trying to create trouble in the international community can obtain and use deep-fake Technology at any time to expand their participation and profoundly influence the situation of international relations.
Deepfakes are likely to become a catalyst for international terrorist activities. Early terrorist organizations lacked the resources to produce false but credible audio and video. Now, they can quickly create such works with the maturity of deep-fake Technology. Terrorist organizations use deep-fake videos or audio to make inflammatory remarks or engage in provocative actions against opponents, including government officials, to maximize the impact on their target audiences, which are favorable to terrorist organizations.
Deep fakes affect news and public opinion
Secondly, deepfakes bring new challenges to the image of the country. On the one hand, deepfakes may be used to discredit the image of the country. At the end of August 2022, the Stanford University Internet Observatory in the United States released a report revealing that the United States used social media to manipulate global public opinion. On the other hand, deepfakes may defame public figures through fake news. As some experts predict, in the next few years, computers will be able to quickly generate convincing, fake audio and video information, which will take fake news to a whole new level. Deep phony Technology may be used to defame public figures through so-called fake news.
Deep Fake violates the public interest
Disruptive Technology has shifted from traditional mysterious technical schools to popularizing technical open-source codes, facing the risk of low-threshold diffusion. For example, the coding algorithm of deepfake Technology has been open-sourced on technical exchange websites such as GitHub; a more convenient way is to download related application software for free through the mobile app store and carry out deep fake activities. Civilians who obtain channels and application scenarios see the trend of deep phony Technology as a threat to social security and economics and infringe upon citizens’ legitimate rights and interests.
Deep-fake Technology infringes upon citizens’ portrait rights and reputation rights. Commercial and entertaining deep-fake software developed by many technology companies can easily integrate facial expressions, technical movements, and sounds. This widespread and socially entertaining use makes it possible to carry out large-scale, low-cost spoofs and even slander against individuals. The most common illegal use of deep-fake Technology is to “transplant” the faces of some well-known singers, movie stars, and other public figures onto porn stars, forge pornographic films for illegal profit, or forge spoof videos, which constitutes a severe infringement on personal reputation and portrait rights.
Policies and Governance of Deep Fake Technology in Various Countries
The United States: bottom-up governance path
In 2019, the U.S. intelligence community released a research report titled “2019 Global Threat Assessment,” which believes that profound fake Technology has threatened U.S. national security. Hostile forces and strategic competitors are likely to attempt to use “deep fake” technology or similar machine learning technology. The aim is to create highly credible but completely false pictures, audio, and video materials to strengthen the influence and infiltration campaign against the United States and its allies and partners.
As the first country to have deep fake Technology on a large scale, the United States has taken targeted response measures to a series of problems caused by this Technology, starting with grassroots participation and integrating all parties to form a bottom-up governance path (see Table 1).
Since the popularity, breadth of influence, and depth of damage of “deep fake” face-changing in the E.U. are far lower than those in the United States, unlike the United States, which has formulated targeted laws, the E.U. has included “deepfakes” in the category of “false information” for unified management.
On April 26, 2018, the European Commission announced “Tackling online disinformation: a European Approach. “The Commission is the governance body, playing a top-down overall role and coordinating the participation of E.U. member states in the governance of “deepfakes.”
China, Multi-party Joint Governance Path
Deep-fake Technology has attracted people’s attention in China due to the face-changing application “ZAO” in 2019. Z.A.O. made the following provisions on user personal information in the initial version of the user agreement: “Ensure that the portrait rights holder agrees to grant “Z.A.O.” and its affiliates completely free, irrevocable, permanent, and sub-licensable. In addition to that, re-licensable rights worldwide, including but not limited to the portrait rights of you or the portrait rights holder contained in portrait materials such as face photos, pictures, and video materials, and the use of Technology to make formal changes to your or the portrait rights holder’s portrait. “However, with the surge in the number of Z.A.O. users, the public has been widely concerned and questioned about Z.A.O.’s forced users to enter facial data, excessive grabbing, and wanton abuse.
Face-Changing
There has been no political public opinion incident of “deep fake” face-changing in my country, and the negative social impact has become prominent. The response measures are still being explored, but a governance path with the joint participation of multiple parties under the government’s guidance has been presented.
At present, many regions and countries, including China, the United States, and the European Union, have recognized the challenges and threats brought by deepfake. It has taken actions from the research and development of new technologies, strengthening the supervision of information release platforms, and social legislation to curb the various threats brought by malicious deepfakes. However, the cross-border nature of digital activities also determines that cooperation between countries is indispensable to solving the problems brought by deepfakes. The international community must advocate the concepts of comprehensive security, cooperative security, and common security based on mutual respect and mutual trust and jointly create a community of shared future for human security in the digital age.
The International Institute of Technology and Economics (I.I.T.E.) was established in November 1985. It is a non-profit research institution affiliated with the Development Research Center of the State Council. Its main functions are to study central policy and strategic and forward-looking issues in my country’s economic, scientific, technological, and social development, track and analyze the world’s scientific, technological, and economic development trends, and provide decision-making consulting services to the central government and relevant ministries and commissions.
