"The irony is that humans can't surpass it no matter how fast they are." This is what singer Stefanie Sun expressed on the MAKE MUSIC website blog about her AI clones all over the Internet.
When everyone was still worried about whether AIGC would make them unemployed, singer Stefanie Sun became the "first person" facing "unemployment" because of the popularity of "AI Stefanie Sun". As more and more netizens made and uploaded related videos , "AI Faye Wong", "AI Jay Chou" and so on have also appeared, but behind the many AI singers "singing as they want", it has also aroused concerns in the security industry about whether the content is compliant and whether the copyright is infringing.
With the continuous development of AIGC technology in recent years, the output of content no longer relies solely on manual work, and machines have gradually taken over this part of the work. Compared with traditional content, today's Internet content is richer and more diverse. However, this also brings many difficulties to content review and management. According to Li Haoze, a security expert from Tencent, AIGC can continuously generate massive content, which makes content security and content protection more difficult; the continuous maturity of synthesis technology has also brought about the production and dissemination of false and wrong content; in addition, the issue of copyright protection has also become an issue. As for the controversial part of artificial intelligence-generated content, whether AIGC content has copyright and how to define infringement has become a difficult problem.
It can be seen that the security challenges brought by AIGC are all-round. For the governance of AIGC content, it is not only necessary to pay attention to whether the generated content violates laws and regulations, but also to have awareness of content security and corresponding protection measures in the entire link.
In response to these problems, Tencent Security has AIGC full-link content security solutions, which can help enterprises cope with challenges more efficiently. This solution can help enterprises solve AIGC content risk control problems through three levels of risk control measures. The first is to deploy corresponding risk control strategies on the input side to judge whether the input instructions violate the regulations, and avoid the output of illegal content from the source.
Then add shielding measures. For example, when an input violation entry triggers the deployed risk control strategy, the entry will be fed back to the customer and prompt that there is a problem with the question method or non-compliance, etc., so as to avoid spit out non-compliance messages.
Corresponding risk control will also be carried out at the output end. Black samples can be marked and saved to the sample library, and associated features can be matched and intercepted at the output end to prevent the model from generating non-compliant content.
In fact, machines rely on input instructions and model training to generate corresponding content. To some extent, "human" as the sender bears a considerable part of the responsibility, and AIGC technology is equivalent to a medium. Therefore, the governance of content security is not only about "governing affairs", but also "governing people", which requires the linkage of the entire industry.
Li Haoze believes that since there are still no relevant laws and regulations specifically for the supervision of AIGC-produced content, the governance of non-bottom-line content requires companies to consciously undertake self-inspection and management responsibilities, and the industry also needs to establish stricter regulations than supervision . Management standards to ensure the ecological quality of content.
The platform side needs to establish a more complete risk control strategy, upgrade from single-point risk control to comprehensive risk control, and flexibly apply combined strategies in the comprehensive risk control to achieve accurate identification and disposal of technology-generated content. Relevant manufacturers also need to constantly learn new risk points and iterate related risk control models to deal with the emergence of more new content risks.
For individual users , experts suggest that they should further enhance their security awareness, keep their eyes open in the face of massive Internet information, carefully screen out false and wrong information brought by AIGC, and avoid being deceived.
"Things about going to the cloud"
The column launched by Nanfang Daily, Nanfang + and Tencent Security plans to better let the public understand the importance of network security and find a suitable solution for them by explaining in-depth explanations of network security issues in simple terms.
Reporter: Ye Dan