news

recently, the cyberspace administration of china issued the "measures for the identification of synthetic content generated by artificial intelligence (draft for comments)", which provides a standard for the identification of ai-generated content. in layman's terms, ai-generated content will be forced to "reveal its true identity" in the future. it is reported that this is also the first national standard for the identification of synthetic content generated by artificial intelligence in the world.

from text to text, to pictures to cross-modal interaction, the evolving ai generation technology is greatly empowering content production. whether it is writing a piece of news, polishing a novel, or commanding a drawing, or writing a song intelligently, ai has helped creators to save the time of thinking hard, significantly improving production efficiency. however, the dividends of technology are also being used by some people with ulterior motives. some mass-produce fake science, and use gimmicks such as calcium supplementation to strengthen bones, spleen and stomach to fool the elderly; some make up fake news at will, and words such as "breaking", "heavy", and "must see" stimulate the eyes of netizens; some change their voices and faces to pretend to be celebrities, and fake ghosts are rampant to cheat for love and money... all these have seriously polluted the online content ecology, and many are directly suspected of illegal crimes.

"content garbage" has been born with the internet and has been repeatedly beaten but never eliminated. however, with the help of ai, garbage fermentation has been upgraded. specifically, the "content garbage" generated by ai is extremely confusing. by learning from the training data, the ai ​​big model can easily grasp the basic characteristics and forms of different content such as text, images, and videos. although the content generated by ai was relatively rough at the beginning, with the continuous iteration of technology, it has bid farewell to the "fake at first glance" stage and entered a state where it is difficult to distinguish the authenticity with the naked eye. second, the cost of ai counterfeiting has been greatly reduced. if counterfeiting in the past still required multiple cost investments, then with the application of the latest ai big model, there is almost no need for human participation throughout the process, and a variety of content such as text, pictures, and videos can be produced quickly and in large quantities, "at your disposal". if this is left unchecked, the consequences will be disastrous, and it is time to strictly govern it.

in contrast, the current "measures" force ai-generated content to "reveal its true identity", which is equivalent to a source of governance. on the one hand, the current "measures" almost covers all current ai-generated content types, such as intelligent dialogue, synthetic voice, face generation, immersive virtual scenes, etc. on the other hand, the current "measures" have put forward detailed regulations on the subject of identification obligations, scope of application, identification type, usage scenarios and deployment methods, and also provide identification examples. at present, the "measures" are still soliciting public opinions, and ai creators, internet platforms, academia and other relevant entities need to work together to improve the top-level design and let the "small logo" play a big role in the governance of the internet content ecosystem.

"preparation is the key to success", and technology governance and technology development must go hand in hand. standardized labeling will make technology more powerful rather than "helping the tiger", and only then can the ai ​​industry develop steadily and achieve long-term success.

source: beijing daily client

process editor: u072