Our reporter Li Pengda
On February 8, in the women's freestyle skiing platform final at the Beijing Winter Olympics, Chinese athlete Gu Ailing won the championship with her outstanding performance in the last jump. The AI synthesis anchor "Xiao Cong" explained the exciting moment of winning the championship in sign language.
As the world's first sign language AI synthesis anchor, "Xiao Cong" uses deep synthesis technology to bring great convenience to the hearing impaired to obtain information.
As a new application in the field of artificial intelligence, deep synthesis technology, represented by deep learning and virtual reality, produces text, images, audio, and video with generative synthesis algorithms, which have attracted wide attention on social media platforms because of their strong entertainment.
Relevant research shows that on mainstream audio and video websites and social media platforms at home and abroad, the number of newly released deep synthetic videos in 2021 will increase by more than 10 times compared with 2017.
But mass adoption has also led to frequent abuses.
Not long ago, the Cyberspace Administration of China issued the Regulations on the Administration of Deep Synthesis of Internet Information Services (Draft for Comment), which stipulates specific provisions on the use, marking, scope of use and abuse penalties of deep synthetic content.
The industry believes that the imminent introduction of new management regulations means that deep synthesis will usher in a critical period of standardized development.
Abundant applications drive the rapid development of the industry
The "Top Ten Trends in Deep Synthesis Report (2022)" jointly released by the Artificial Intelligence Research Institute of Tsinghua University and the National Industrial Information Security Development Research Center recently pointed out that since 2017, deep synthetic content has been created and disseminated in large quantities, and the number has increased rapidly year by year.
The continuous maturity of technology is an important reason for the rapid growth of deep synthetic content.
Since 2017, the number of newly published papers and open source projects in the field of deep synthesis has grown at a rate of 30% per year.
"Research papers continue to increase, open source technical tools and a large number of representative methods emerge in a concentrated manner, making the effect of deep synthesis content more realistic and production more efficient." Tian Tian, CEO of Beijing Ruilai Smart Technology Co., Ltd. told reporters that throughout the development of the computer industry, Open source projects have become a powerful force in promoting industrial progress, and deep synthesis is favored in the open source community, and will continue to promote the implementation of this technology in the industry.
By upgrading traditional content production methods, deep synthesis has been continuously enriched in film and television production, advertising marketing, social entertainment and other fields, including AI synthesis of anchors, virtual idols, restoration of historical old photos, etc.
In 2021, the virtual idol Luo Tianyi will appear on the CCTV Spring Festival Gala, and in 2022, she will appear on the stage of the Lantern Festival Gala again after 10 years of "debut".
According to public data, from June 2020 to May 2021, a total of 32,412 virtual anchors started broadcasting on Bilibili, a year-on-year increase of 40%.
At the same time, more and more enterprises have begun to use deep synthesis technology to provide public-facing products and services, covering image, video, audio, text and other fields.
Speech synthesis has become an important part of human-computer interaction and is used in scenarios such as intelligent customer service, voice navigation, audiobooks, and voice assistants. The deep synthesis of the form shows great creative efficiency and potential in news reporting, poetry creation, chat and question and answer, etc.
In addition, the proposal of new business thinking such as "metaverse" also provides broader application scenarios for deep synthesis.
"Deep synthesis will redefine the virtual digital space. From the perspective of communication sociology, a new human survival scene will be launched with deep synthesis technology as the cornerstone." said Chen Changfeng, Executive Deputy Dean of the School of Journalism and Communication, Tsinghua University.
Risk-increasing detection technology continues to update
While deep synthesis stimulates new forms of content creation, it also brings new threats and challenges.
The "Top Ten Trends in Deep Synthesis Report (2022)" analyzes that deep forgery affects the record of truth in news, and the difficulty of screening false content also reduces the effectiveness of fact-checking.
In major social emergencies, deep synthesis technology may be used to manipulate public opinion and use social media to ferment false information in a short period of time.
Traditional biometric-based identification methods are becoming increasingly difficult to function as negative risks increase and synthetic quality improves.
"At present, the automatic identification of deep synthetic content mainly relies on artificial intelligence technology." Ren Kui, dean of the School of Cyberspace Security of Zhejiang University, introduced that training artificial intelligence models requires a large amount of real and fake data, and face and audio data are highly sensitive. Personal information is difficult to obtain, and forged data also includes data synthesized by various methods, all of which bring challenges to building automated detection capabilities.
With the endless emergence of new forgery methods and the structural defects of detection algorithms, anti-deep forgery detection technology faces "strong confrontation".
Tian Tian explained that this is similar to a "cat-and-mouse game". In-depth synthesis and detection will self-evolve in the process of continuous learning of attack and defense, avoiding the previous generation of confrontation technology. Therefore, detection technology needs to be continuously updated and iteratively optimized.
At present, both academia and industry have invested a lot of research on anti-deep counterfeiting detection. Organizations such as Google and Microsoft have launched methods or products for deep synthetic video authentication.
In China, DeepReal, a deep forgery content detection platform launched by Relais Intelligence, has industrial-grade detection performance and the ability to detect changes in the real network environment.
"Deep forgery detection faces continuous attack and defense and games. In the future, it is necessary to integrate multi-modal content forensic analysis, digital watermark-based traceability technology and other capabilities to achieve accurate identification." Director of the Basic Theory Research Center of the Institute of Artificial Intelligence, Tsinghua University Zhu Jun said.
Build a multi-dimensional governance mechanism
In recent years, in response to the problems caused by the malicious use of deep synthesis technology, countries around the world have issued management laws and regulations to explore the governance path of deep synthesis.
The European Union has incorporated deep synthesis into existing legal frameworks such as the General Data Protection Regulation (GDPR). Germany, Singapore, the United Kingdom, South Korea and other countries have laws and regulations applicable to the trial of deep synthesis technology-related crimes.
my country is actively exploring the establishment of an effective governance mechanism.
Since November 2019, documents such as the "Regulations on the Administration of Online Audio and Video Information Services", "Regulations on the Ecological Governance of Network Information Content", and "Regulations on the Management of Internet Information Service Algorithms Recommendation" have all put forward different levels of supervision on the generation of synthetic content. Require.
Wu Hequan, an academician of the Chinese Academy of Engineering, believes that the governance of deep synthesis cannot be "one size fits all", and it is necessary to continuously develop technology to avoid hindering its positive application and innovation.
The security problems derived from it need to be solved from the source, and guide the artificial intelligence academia and industry to continuously strengthen technology research and development, and expand research on deep synthesis traceability, deep synthesis identification, etc., to prevent ethical security risks and compliance risks.
To guide the healthy development of deep synthesis technology, it is necessary to explore a multi-dimensional governance mechanism.
Duan Weiwen, Director of the Research Office of Philosophy of Science and Technology, Institute of Philosophy, Chinese Academy of Social Sciences, suggested that systematic and prospective interdisciplinary research on deep synthesis of technical, legal and ethical issues should be strengthened, and targeted governance and high-risk application scenarios should be adopted. Supervision.
Zeng Yi, a researcher at the Institute of Automation of the Chinese Academy of Sciences, advocated the development of self-discipline and autonomy for industry, academia, and research. He said that before laws and regulations have been systemized, the industry itself should strengthen the awareness of theory first, prevent abuse, and strictly prohibit malicious use.
According to Xu Xu, an associate professor at the School of Law of the University of International Business and Economics, the social level should increase publicity and popularization, strengthen citizens' understanding of artificial intelligence technologies such as deep synthesis, and improve the awareness of prevention in the whole society.
Tian Tian has the same opinion on this. He believes that the essential problem of deepfake is lack of transparency. Therefore, it is particularly important to improve the public's awareness of deep synthesis technology. Only by lowering the threshold to the point where all audiences can recognize, discuss and understand this problem under a common framework. When there is a problem, deep synthesis technology can develop healthily and soundly.
Industry experts suggest that all parties should implement the new normative requirements, constantly pursue technological breakthroughs under this premise, develop application scenarios of deep synthesis technology, and form a driving effect on the artificial intelligence industry.