Shulou(Shulou.com)11/24 Report--

Since the release of ChatGPT, the research and application of generative artificial intelligence has been very hot. Compared with previous artificial intelligence technologies, generative artificial intelligence (AIGC) embodies typical characteristics such as strong human-computer interaction and strong corpus dependence, and there are many challenges in technology development and risk control.

On October 18, in order to further promote the healthy development and standardized application of AIGC services, and promote collaborative innovation between colleges and universities and industry, the special event "generative artificial Intelligence Application and Security" was held in Beijing under the guidance of Beijing Science and Technology Association, sponsored by Beijing Software and Information Service Industry Association, and co-sponsored by the artificial Intelligence Security Special Committee of Beijing artificial Intelligence Industry Alliance. Experts and scholars from all walks of life, such as Xintong Yuan, China University of political Science and Law, Tencent and so on, were invited to attend the meeting. At the round-table meeting, Wang Xiang, general manager of Tencent Security content Security, focused on AIGC-related policies, technology trends, security risks and governance issues.

(Wang Xiang, General Manager of Tencent Security content Security, spoke at the special event)

Wang Xiang said that with the formal implementation of the interim measures for the Management of generative artificial Intelligence Services, China has filled the policy gap in the R & D and service of generative artificial intelligence for the first time. This is also the first legal document in the world to regulate generative artificial intelligence. " The measures implement inclusive, prudent, classified and hierarchical supervision of generative artificial intelligence services, and stipulate to promote the orderly opening up of the classification and classification of public data, which will promote the development and application of the industry to be more healthy and standardized, and further stimulate market vitality and social creativity. "

The development of generative artificial intelligence indicates that artificial intelligence is developing from special intelligence to general intelligence and has entered a new stage of development. Relevant data show that China's artificial intelligence industry is booming, the scale of the core industry has reached 500 billion yuan, innovative achievements continue to emerge, the AIGC industry will usher in greater development opportunities. Among them, there are more than 300 AIGC manufacturers, which are mainly divided into application layer, model layer, data layer and computing layer. Wang Xiang mentioned that at present, application layer manufacturers account for a large proportion, mainly focusing on six scenarios, such as marketing, customer service, office, interaction design, digital people, and writing assistants, so companies that can quickly fit in with end users still have a chance to break through. The technical requirements of model layer, data layer and computing layer are relatively high, so the enterprises of these tracks need to focus on effect, cost and service quality in order to achieve greater breakthroughs.

The rapid development of generative artificial intelligence also has many potential risks, such as content security, data security, AI application security and so on. From the point of view of the source of risk, it basically covers the whole link from the training phase of the model, the generation phase to the launch of the application; from the type of risk, issues such as data quality and credibility, countering sample attacks, information security, technology abuse and moral hazard, user privacy and identity security have also been criticized. In addition, because the generating ability of AIGC mainly comes from data and models, and the data source itself may lead to the risk of infringing upon other people's copyright, portrait rights, reputation rights, personal information and other rights, there will be more and more calls on how to protect the intellectual activities of human subjects all over the world.

Wang Xiang mentioned that AIGC track enterprises need to strictly train the full-link content security management from the model to the online operation, ensure the authenticity, harmlessness and availability of the data source, and set up a security team to conduct a comprehensive audit of the generated content through machine audit and manual audit, in order to meet regulatory needs and avoid infringement problems. At the same time, enterprises can also seek access and help from external three-party manufacturers.

To solve these problems, Tencent Security has an AIGC full-link content security solution, including revision service, security expert service, machine audit service and copyright protection service, covering the content security construction of AIGC applications from model training to content generation to ex post operation, which can help enterprises to meet challenges more efficiently.

This scheme can help enterprises solve the problem of risk control of AIGC content through risk control measures at three levels. First of all, the corresponding risk control strategy is deployed on the input side to judge whether the input instruction violates the rules or not, so as to avoid the output of the illegal content from the source. Then take shielding measures, such as when the illegal entry triggers the deployed risk control policy, the entry will be fed back to the customer and prompted that there is a problem with the method, or non-compliance, and so on, in order to prevent non-compliance messages from spitting out. The corresponding risk control will also be carried out on the output side, and the black samples can be marked and saved to the sample database, and the associated features can be matched and intercepted at the output side to prevent the model from generating non-compliant content.

Currently, Tencent Security has implemented a number of scenarios in the field of AIGC content security, providing content security services for enterprises such as Wen Sheng Wen, Wen Sheng Graph, grade video content, long novel text, and so on. Since September 2019, Tencent Security has conducted copyright monitoring services for original videos uploaded by UP on the bilibili platform. It has monitored more than 5 million works, found nearly 30 million infringement chains, initiated nearly a million rights protection, and achieved a success rate of 98%. In addition, relying on the rich copyright material library, Tencent Security successfully helped a certain AIGC create App to eliminate the risk of infringement by comparing the copyright of "generating pictures" with "picture library".

Wang Xiang mentioned, "although the world is actively embracing AIGC, its content security risks will also have an impact on social, economic, cultural and other aspects. How to promote the orderly and healthy development of generative artificial intelligence deserves the common attention of the government, enterprises and other sectors of society." In the future, Tencent will uphold the concept of Tech for Good, create secure and reliable AIGC technology and applications, and help achieve high-quality, healthy and sustainable development in the AIGC era. "

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.