comment: add prominent labels to prevent ai-generated content from being mixed with fakes
2024-09-15
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
of the massive amount of images, texts, and videos we see online every day, how many are generated and synthesized by artificial intelligence? how can we accurately distinguish ai articles, ai images, and ai videos?
on september 14, the cyberspace administration of china issued the "measures for identifying synthetic content generated by artificial intelligence (draft for comments)". the "draft for comments" proposes to stipulate that the generated synthetic services provided by service providers, including text, audio, pictures, videos, virtual scenes, etc., should all add prominent prompt labels in appropriate locations; when service providers provide methods for downloading, copying, and exporting generated synthetic content, they should ensure that the files contain explicit labels that meet the requirements.
this regulation is very realistic, targeted and necessary. in recent years, ai deep synthesis technology has been widely used in many fields, from image, text and video generation to online question and answer, and various applications are emerging in an endless stream. there is no doubt that ai tools can make our content production and knowledge dissemination more efficient, and as ai's learning ability becomes stronger and stronger, its content "production" ability will even exceed that of humans in the future.
on the other hand, the effects of deep synthesis content are more realistic and the production is more efficient, which also makes the synthetic content reach the level of "difficult to distinguish between true and false". when the true and false are mixed and difficult to distinguish, seeing is no longer believing. artificial intelligence technology is like a double-edged sword, which will also bring problems such as information distortion and cognitive confusion to society.
in november 2022, the cyberspace administration of china issued the "regulations on the management of deep synthesis of internet information services", which made specific provisions on the purpose, marking, scope of use and abuse penalties of deep synthesis content. the "draft for comments" further clarifies that whether it is text, audio, pictures or videos, "significant prompt marks must be added in appropriate locations." this is a clear requirement for service providers: as long as it is generated by ai tools, whether it is an article or a picture, a significant mark must be added.
this is actually also a reminder to the audience that there is a mixture of good and bad information on the internet. when you see "inside" articles, "explosive" pictures or "shocking" videos, you must first pay attention to whether they are generated by ai, whether you should believe them, and how to distinguish them. you must be more careful.
it is worth noting that as technology gradually becomes more popular, criminals can easily use ai to spread rumors and produce fake news. not long ago, a media outlet exposed an mcn agency that can use ai to generate 4,000 to 7,000 rumor articles every day. these articles look very "real and reasonable" because of the support of ai technology. audiences who are not familiar with internet technology can easily believe this kind of "picture-based and truthful" information content.
according to a media investigation, the cost of using a domestic ai to write an ai article is only 0.00138 yuan, and 1 cent can generate 7 articles. such "no cost, huge profits" can be easily exploited by criminals. false pictures and videos such as "subway attack", "xi'an explosion", and "explosion in a residential building in wuxi, chongqing" have had a bad social impact. in fact, the rumors behind them are mass-produced by ai.
it is not technically difficult to identify synthetic content generated by artificial intelligence. some experts suggest that the platform can fully access the large model system of the ai content generation platform, perform real-time data recognition and comparison, and then add an indelible digital watermark, or text description such as "this image/video is ai-generated".
the draft for comments emphasizes that "no organization or individual shall maliciously delete, tamper with, forge, or conceal the generated synthetic content identification specified in these measures", which is a reminder to ai service providers and platforms that they must not turn a blind eye to the "traffic dividend" brought by ai tools. a clear cyberspace must not allow ai content to be mixed with fakes and blur the boundaries of reality; if any information is identified as a rumor, the platform should immediately mark and clean it up.
the ability of artificial intelligence to generate synthesis exceeds traditional cognition, and also makes us realize that many online articles that look very "scientific" and many online videos that look particularly "real" are not necessarily reliable, so don't rush to believe them. you might as well look for a prominent position in the content to see if there is a sign that says "this image/video is ai-synthesized"?
li qinyu, chief commentator of the paper
(this article is from the paper. for more original information, please download the "the paper" app)
