Video face change, how to manage?

Photo: Visual China

Source: 21tech

Author: Yang Qingqing

Editor: Zhang Weixian

As one of the areas of the artificial intelligence industry, the importance of deep synthesis is penetrating into all aspects of life.

Whether it's video face swapping, speech synthesis, or image restoration, or the virtual digital people who have recently been on fire due to the metacosm, deep synthesis technology is at work. At present, deep synthesis is appearing more and more frequently in social entertainment, film and television production and other fields, and developing diversified digital applications.

In terms of content volume, deep synthesis industry applications are also ushering in leapfrog development.

According to the "Report on Ten Trends of Deep Synthesis (2022)" (hereinafter referred to as the "Report") jointly released by the Institute of Artificial Intelligence of Tsinghua University, the National Industrial Information Security Development Research Center and other institutions, the number of new releases of deep synthesis videos in the Internet has soared from 1680 in 2017 to 24317 in 2021, and the number of likes/likes of deep synthesis videos in the Internet has also jumped from 2.4974 million times in 2017 to 316 million times in 2021.

As the industry booms, the pace of industry-related compliance is also accelerating. On March 1, the Provisions on the Administration of Internet Information Service Algorithm Recommendations, jointly promulgated by the State Internet Information Office, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation, came into effect, which clearly requires that synthetic false news information must not be generated.

On January 28, 2022, the State Internet Information Office also promulgated the Provisions on the Administration of Deep Synthesis of Internet Information Services (Draft for Solicitation of Comments), which aims to promote the rational and effective use of deep synthesis technology in accordance with the law and standardize the development of internet information service deep synthesis activities.

In an interview with the media, including 21st Century Business Herald, Duan Weiwen, a researcher at the Institute of Philosophy of the Chinese Academy of Social Sciences and director of the Research Office of philosophy of science and technology, said that deep synthesis technology involves cognitive security, so it is necessary to enhance the awareness of social cognitive security and enhance the public's digital literacy. At the same time, government departments, enterprises and all walks of life need to actively carry out dialogues and norms on relevant laws, ethics and governance, so as to achieve a dynamic balance between innovation and governance.

Apps speed up

Facial replacement, expression manipulation, speech synthesis... With the development and maturity of technology, more and more deeply synthesized content has appeared on the Internet.

The "Report" shows that according to incomplete statistics, the number of deeply synthesized content released by creators on the Internet platform is showing rapid growth. Taking video as an example, the number of newly released deeply synthesized videos in 2021 has increased by more than 10 times compared with 2017.

Not only has the amount of content itself exploded, but the attention of deeply synthesized content is also growing exponentially. The report pointed out that taking the "likes/likes" data of videos as an example, since 2017, the data has shown an exponential significant increase, and the number of newly released deeply synthesized videos in 2021 has exceeded 300 million.

In addition, the report selected more than 4,000 interactive videos of deeply synthesized videos, and analyzed the relationship between the interactive data of "playback, likes/likes, favorites, retweets, and comments" and found that on average, every 1,000 plays can generate about 8.92 likes/likes, and about 3.51 forwards will be generated, thus a new round of dissemination of deeply synthesized content.

In addition to the content level, at the technical level, the field of deep synthesis is also developing rapidly. The report shows that the number of papers published in the field of deep synthesis reached 4559 in 2021, a significant increase from 1012 in 2017. Among the papers in these deep synthesis fields, there are technical studies on the synthesis of different modes such as images, video, audio, and text, of which the research on image generative models accounts for 64%.

At the same time, since 2017, the number of open source projects released in the field of deep synthesis has also increased year by year. By 2021, the number of open source projects related to 16 keywords such as "GAN" has reached 23,030 in GitHub, an increase of 179% compared to 8,250 in 2017. The continuous high popularity of deep synthesis in the open source community will continue to promote the development of this technology and the landing in the industry.

In fact, the current application scenarios of deep synthesis have been continuously enriched, whether it is sign language AI generation anchors, virtual idols or repair of old photos, AI face changes, etc., with the continuous improvement of technological maturity, deep synthesis is rapidly entering the public vision, and more and more enterprises and institutions have begun to use deep synthesis technology to provide products and services for the public.

This also greatly lowers the threshold for deep synthesis. Tian Tian, CEO of Ruilai Wisdom, said that taking the virtual digital people who have been popular since the second half of last year as an example, if there is no deep synthesis technology, the cost of virtual images is very high and requires strong professionalism. "However, based on deep synthesis technology, AI automatic learning can reduce costs and thresholds, so that more people can participate in content generation and promote the development of digital space."

Compliance governance

With the continuous development of the deep synthesis industry, the technical threshold of content production continues to be lowered or even the "civilianization" of technology is achieved, and the related content risks continue to increase.

In fact, through the deep synthesis technology to create false video, false audio framed, defamation, fraud and other illegal acts and cases have been seen frequently, the number of deep forged content continues to increase, the harm continues to increase, therefore, the governance of the deep synthesis industry is also an urgent problem to be solved.

"At present, the identification demand for deep synthesis is growing rapidly, but the difficulty of identification is also increasing rapidly." Tian Tian admitted to the 21st Century Business Herald reporter that the industry challenge is mainly reflected in the rapid improvement of deep synthesis fidelity, and it is even difficult to distinguish the true and false boundaries, in addition, the latest deep forgery technology has strong confrontation, naturally targeting detection methods to counter attack, so that detection methods fail.

In order to promote the development of deep forgery detection technology, special research plans and related academic competitions have been launched at home and abroad. At present, academia and industry have made a lot of investment in the research and development of deep synthesis identification and detection technology, including Meta, Google, Microsoft and other institutions have launched video certification methods or products, and institutions and enterprises in China, including Tsinghua University, University of Science and Technology of China and other institutions and enterprises, have also built a face synthesis detection platform and released targeted detection products.

From the perspective of regulatory mechanisms, domestic and foreign countries are also continuing to explore. As early as 2018, the United States proposed the "Malicious Forgery Ban Act of 2018", and in 2019, it continued to issue the "Deep Forgery Responsibility Act", and continued to strike hard at deep forgery. Since 2018, the European Union has also successively launched the "Application of Online False Information: European Solution", "General Data Protection Regulation (GDPR)", "Code of Conduct Against False Information", etc., to combat violations related to deep forgery.

The mainland is also actively exploring the construction of an effective governance mechanism. Since 2019, including the "Provisions on the Administration of Online Audio and Video Information Services", the "Provisions on the Ecological Governance of Online Information Content", and the "Provisions on the Recommendation and Management of Internet Information Service Algorithms", all of which have restricted the violations of laws and regulations on deep synthesis, and in January this year, the "Regulations on the Management of Deep Synthesis of Internet Information Services (Draft for Comment)" was promulgated, thus forming special management regulations on the systematic, targeted and operability of the field of deep synthesis.

Duan Weiwen pointed out that the sound governance of deep synthesis requires multiple measures at the same time, not only technical management but also content management, and at the same time needs to enhance the overall social cognition and safety awareness. In addition, there is a need for a prescient and robust governance framework within the industry.

"Now the industry hopes that the government or relevant departments and all walks of life will actively carry out dialogues on legal ethics and governance, the formulation of relevant norms, and the promotion of practices, all of which are aimed at achieving a dynamic balance between innovation and governance, and implementing agile and stable governance, so that technology can better serve digital transformation." Duan Weiwen said.

Editor: Lu Taoran