Administration of AI Generated Content and FiveYear Plan for Development of Robotics Industry
Issue 17, 07 Feb 2022
I. Military and Warfare
Due to Lunar New Year and Beijing Winter Olympics, there was no noteworthy perspective on military and warfare.
II. SinoSync
Administration of AI Generated Content
CAC has floated a draft of provisions on administrations of synthetically generated content using AI (official version here, unofficial translation here). The provisions have been framed under the Cybersecurity Law, Data Security Law, Personal Information Protection Law, Measures for Administration of Internet Information services, etc. State Internet Information Office (国家互联网信息办公室 Guójiā hùliánwǎng xìnxī bàngōngshì) (SIIO) will be responsible for the interpretation of these provisions. The draft targets applications of "deep synthesis" technologies to provide internet information services.
Key terms from the proposal and their meaning:
Deep Synthesis: This term in the proposal refers to technologies that use AI algorithms to generate or manipulate content. In the draft, "deep synthesis" refers to text, images, audio, video, virtual scenes, or other information generated using deep learning artificial intelligence technology.
Deep Synthesis Service Providers: Platforms, companies, and tools, that use deep learning AI to edit content or allow users to edit content. Also includes app stores that are used to distribute applications.
The document explains six key areas (but not exhaustive) where "deep synthesis" is used:
The listed motivation behind the proposed regulations is national security and preserving law and order. Article 6 of the draft regulation states that "deep synthesis services must not be used by any organization or individual to engage in activities that are prohibited by laws and regulations, such as those endangering national security, undermining social stability, disrupting social order, or violating the lawful rights and interests of others…." (第六条 任何组织和个人不得利用深度合成服务从事危害国家安全、破坏社会稳定、扰乱社会秩序、侵犯他人合法权益等法律法规禁止的活动，不得制作、复制、发布、传播含有煽动颠覆国家政权).
This includes any content that:
Can subvert state sovereignty
Endanger national security and social stability
Is pornography
Has fake information
Infringe upon intellectual property rights
Distorts reputation of someone
As per the draft, internet platforms and providers of deep synthesis services have the following responsibilities:
Register with CAC within 10 days of starting to provide such services.
Establish information security and management systems (like algorithm review mechanisms, user registration, information content management, etc.)
Conduct real-name identity verification for users.
Strengthen content management (review data inputs by the user, employ measures to address illegal and negative information, etc.)
Strengthen the management of technologies (periodically reviewing, assessing, and testing algorithmic mechanisms, etc.)
Strengthen the management of training data
Alert the public about synthetically generated content and stop transmission if it is not labeled.
Establish mechanisms for dispelling rumors.
Mechanism to manage public complaints.
My thoughts on the draft:
More than deepfakes
Although this draft is being primarily read as targeting deepfakes, the "deep synthesis" technologies can be used for a range of different purposes. The draft itself mentions six broad areas that include not only image and voice manipulation but also text editing and 3D reconstruction.
Responsibility to manage content shouldered on platforms and service providers
The burden of management of synthetic content is placed on internet platforms and service providers. This includes companies who make tools used to generate and edit such content. The responsibility to track, verify, and ensure legality, is also on service providers. Moreover, the list of applications mentioned in the draft is not exhaustive. This means the interpretation of applications using deep learning AI tools might be subject to interpretation.
Significance for military
Generated content can be used for a range of purposes apart from deepfakes. The future warfare and battlefield conceptualized by the PLA researchers and officers will need generated content using deep learning AI.
For example, to make "battlefield metaverse" a reality, PLA will need superior capabilities to generate and manage synthetic content. This includes 3D reconstruction, editing biometric attributes, face generation, gesture manipulation, etc.
Hence regulation of such technologies will nudge platforms and providers towards building a resilient digital infrastructure suitable for military adaptation.
Significance for national security
Deepfakes can be used to target leaders, spread misinformation to create social discontent, destabilize the government, discredit a leader, etc. Hence, it can be a potent tool for offensive and defensive information operations. Hence, it is not surprising for Beijing to try and bring generated content under regulation. This is continuing with the trend of bringing more accountability to internet platforms (like 2019 rules banning deep learning to produce fake news videos and audios).
Also, anything deemed politically sensitive or otherwise unfit will come under the scanner.
On the other hand, provisions to label the generated content ensures that users will know when they are using or viewing synthetically generated or edited content. Hence preventing the spread of misinformation. For example, if you are watching a video of a viral speech, and it has been altered, the platforms will have to either take down the video or label it as being manipulated.
Data Security and Privacy
Draft mandates that users need to obtain consent of a person before altering biometric features. This is consistent with other laws and guidelines emphasizing data security and privacy. The Chinese government views data security and privacy as a national security matter.
Also, refer to this thread:
Five-year Plan for Development of Robotics Industry
Ministry of Industry and Information Technology (MIIT), China along with more than ten other departments and ministries has released "Five-year Plan for Development of Robotics Industry" (“十四五”机器人产业发展规划 “Shísìwǔ” jīqìrén chǎnyè fāzhǎn guīhuà) for the 14th Five-year Plan period (2021-2025). This is second five year plan for the development of robotics industry by Chinese government. The previous one was released during 13th Five-year planned period (2016-2020).
The Plan is for development of industrial robotics and focus in on making China a world leader in robotics innovation and high-end manufacturing sector by 2025. By 2035, the goals is to increase comprehensive strength of robotics industry to an international level and make robotics an important component of economic development, people's life, and social governance.
Specific goals to be achieved by 2025 are:
Breakthrough in core robotics technologies and high-end industrial products (一批机器人核心技术和高端产品取得突破)
Comprehensive indicator of whole industry has reached international level (整机综合指标达到国际先进水平)
Achieve international level standards in performance and reliability of for key components and products (关键零部件性能和可靠性达到国际同类产品水平)
The average annual growth rate of the robot industry's operating income exceeds 20% (机器人产业营业收入年均增速超过20)
Build a group of internationally competitive and professional "little giants" with strong innovation capability and growth potential (形成一批具有国际竞争力的领军企业及一大批创新能力强、成长性好的专精特新“小巨人”企业)
Build 3 to 5 internationally influential industrial clusters (建成3-5个有国际影响力的产业集群)
Double density of manufacturing robots(制造业机器人密度实现翻番)
The Plan outlines five main tasks:
Improve industrial innovation (提高产业创新能力)
Consolidate foundation of industrial development (夯实产业发展基础)
Increase supply for high end products (增加高端产品供给)
Expand depth and breadth of applications (拓展应用深度广度)
Optimise the industrial organisation structure (优化产业组织结构)
The plan also has proposed safeguard measures to promote smooth implementation:
Strengthen planning and coordination (强化统筹协调推进)
Increase financial support (强化统筹协调推进)
Create a favourable market environment (营造良好市场环境)
Improve talent protection (健全人才保障体系)
Why focus on robotics?
First, primarily focus is on industrial robotics, the ones used to automate manufacturing process. This directly resonates with Chinese government's aim to move manufacturing industry to high-end manufacturing. This is in continuation and supplement to others plans like Made in China 2025.
Second, this plan is linked to the intelligent manufacturing (smart manufacturing) plan (Read Issue 15 of China Tech Dispatch) with similar goals to elevate manufacturing industry in China. Robotics is part and parcel of the automation required to achieve intelligent manufacturing.
Third, the areas focussed in the Plan also indicate areas the government is concerned about. For example, there is mention of robots for healthcare, elderly care, security, epidemic prevention, etc. China's population is aging fast and The aging population is one of the areas of concern for the government. According to China's seventh population census, population of people aged 60 was 264 million(18.7%) and 65+ was over or 191 million (13.5%). By 2050, China's aging population is expected to reach nearly 500 million. Hence, elderly care is propping up in many policies and plan of the government.
Fourth, Chinese officials consider robots as am “important drivers of emerging technologies” as also quoted in ChinaDaily. Emerging technologies like Artificial Intelligence and Big Data now feature in many plans of Chinese government aimed to achieve goals of “Digital China.”
Chinese companies import industrial robots primarily from Japan, Korea, and Europe. According to International Federation of Robotics, installations of robots from abroad in China grew by 24% in 2020.
Additional Reading:
III. InfoBytes
Deep Learning and Synthetic Content Generation
Deep learning is a technique in machine learning (branch of AI) that is used to teach machines tasks that humans can perform naturally. This includes generating content like text, images, music, etc. Deep learning uses "learn-by-example" method to teach machines perform tasks like categorizing, labelling, extraction, transformation of data, etc. The most important advantage of deep learning in content generation is accuracy and speed with which machines can produce content. I have covered one of the methods used to train machines in previous issue of China Tech Dispatch. Read Smart Brain, Downgrading Operations, and GANs in Future Wars.
AI is being used to create a lot of content like images, texts, videos, music and other things. For example, this Narrative Device app developed by PhD researcher Rodolfo Ocampo generates a story based on two words you feed into the app. Look at the story this app created for me.
Another example is of Random Image Generator website. This website generates a random image of a person which does not exist. Look at the image of woman below, this very realistic image is of a woman who does not exist.
Another one uses AI to generate image for text description.
Such content generation has become a cause of concern because of "deepfakes", hyper realistic media or content generated using AI. For example, this following speech of former US President Obama generated by researchers of University of Washington.
Barrack Obama never gave the speech in this video, it was created using his audio from previous speeches, studying his style, and past images. While amazing, this technology can also be misused. Deepfakes and synthetically generated content can have national security implications since it can be used to spread misinformation among other things. Unless you have highly specialised tools, you cannot easily tell difference between real or fake. Hence, a cause of concern for governments.
(Note: The examples of AI applications above are just used to show how AI is used to create content. It doesn't imply that these applications are malicious.)
Megha Pardhi is a Research Analyst at The Takshashila Institution. She tweets at @pardhimegha21.