Baidu Netdisk Suspected of Conducting Human Content Reviews

Are Machines Fully Qualified for Censorship? Is Privacy at Risk?

Edited by Lin Jiang and published at Southern Metropolis Daily, Aug. 21, 2022

Baidu Netdisk was accused of going through a human review of users' cloud storage images on August 19, 2022. In a short video, a man claiming to be a part-time examiner displayed blurred images purportedly of users. Baidu immediately issued a denial.

The Public's main concern is whether other people are viewing their private items on Netdisk. Others questioned the platform's review process.

According to technical experts, it was "unlikely" that a machine would completely analyze anything a person supplied given present AI capabilities. However, lawyers clarify that when a machine can't tell, human inspection is not prohibited. It is suggested that Baidu should at least include the review procedure in the user agreement.

Baidu Netdisk: There is no "human review of user photos"

A man who claimed to have been hired as a part-time online examiner for Baidu Netdisk posted a short video online, with his admittance from the Qingtuan job recruitment platform. Entitled "Why do girls enjoy snapping so many selfies? I'm going blind to make money", the video featured many blurred images that were allegedly collected from users' cloud storage.

On August 18, the screenshot of the video was posted online by writer Bian Xiang. He wrote on his blog, "I never imagined that real people were there behind Baidu Netdisk? I always assumed that it was all reviewed by machines. Can users' photos be saved and posted online by the examiners?" The post quickly sparked a heated debate, with related topics topping the list of popular searches.

Another similar short video was also discovered. A female shared a list of folders that "can be seen" by a Baidu Netdisk examiner. Both videos revealed what seemed to be text messages from the Qingtuan job platform for part-time job admittance.

In less than two hours, Baidu Netdisk officially stated that "human review of user photos does not exist”, no matter the technology or product terms. According to Baidu, who had already filed a police report, the material from the online video was made up.

Bian Xiang was contacted by Qingtuan late at night in a private chat on Weibo, who informed him that "there are no similar job posts on record". These two videos were taken down from Tiktok the following morning, and it is now impossible to find the bloggers using their screen names.

On September 19, Qingtuan News Agency stated to Nandu reporters that there were no C2C positions on the platform. Self-inspection had already been done the previous evening, and the blogger is thought to have made up that information. Qingtuan further disclosed that it intended to issue a statement as well.

The screenshot was a false rumor, according to the second statement from Baidu Netdisk, which was released on August 20. Additionally, it "did not entrust any agency with the recruitment of part-time personnel." The privacy and data of users are adequately protected by the standard, stringent privacy management mechanism, and security defense system, said Baidu Netdisk.

No employment information was associated with a keyword like "Netdisk review" in Qingtuan, as Nandu reporters found. According to the 2018 internship recruitment of examiners for Baidu Netdisk on a WeChat account "Internship Net”, the work content is primarily to mark information accurately and to assist the technical department in improving the reviewing strategy.

How does a cloud storage platform conduct the content review?

The requirements for content review in China can be seen in several documents issued by central ministries.

According to the Standard of Management in Online Short Video Platforms, all short videos should be approved before broadcasting, including the title, introduction, bullet screen, comments, and other content.

Website platforms must strengthen the human review system by broadening its reach and enhancing its standards, as stipulated by Opinions on Further Compacting the Main Responsibility of Website Platform Information Content Management in 2021.

A platform has the right and duty to monitor the content that users upload, according to Zheng Jingjing, a lawyer from Jingheng Law Firm. It is also mentioned by Xiong Dingzhong, chief partner of Qinglv Law Firm, that companies offering cloud storage space may risk facing legal repercussions for allowing the storage of illegal and undesirable information.

Platforms have increased their investment in content review over the past few years. Most platforms adopt a combination of automated and human assessment. Artificial intelligence (AI) will examine, mark, and filter user-uploaded content by comparing it to a database of infractions. If judging is challenging, a risk alert will be displayed to aid human review.

Can the content review be carried out entirely without the aid of a human at the current level of technology?

There are several thresholds for information in different formats. The easiest format for machine evaluation is text, according to Peng Gen, general manager of Beijing Hanhua Feitian Xinan Technology Corporation. A certain level of complexity exists with photos. Given that a video is composed of many pictures, the risk is greatly increased because the machine typically extracts some of the pictures for inspection.

Platforms have the technical capacity to see each file that a user uploads, Peng Gen points out. It was "unlikely" that users’ entire upload would be completely reviewed by a machine given current AI capabilities.

He gave the example of the job of a "porn cleaner", which picks out pornographic movies online and has not been entirely replaced by technology over time. "I believe that AI for cloud storage shouldn't be stronger. There must be a human part in it, but only the company might be able to say when."

Expert: Platforms should inform users of the content review process

It's worth pointing out that even if a platform uses human review, it doesn't necessarily violate users' privacy, as Xiong believes. For example, user can ask for unblocking due to AI misjudgment. In such cases, human review is reasonable because it is based on the user's request. Additionally, cloud files are no longer covered by privacy protection if they have been shared with others.

According to Xiong, the present review in cloud storage will not violate users' privacy; rather, it is more of a right-to-know problem, and Baidu should at the very least include a description of the entire review process in the user agreement. "How do they conduct the content review? What statutes and rules apply here? When is human intervention necessary? It must be made explicit."

Nandu reporter searched for the keyword "review" in Baidu Netdisk's privacy policy and user agreement and found no pertinent text.

Zheng Jingjing also agrees that the platform should inform the content review method in the user agreement or website, including the purpose, type, and method of dealing with user information. The user's right to know is violated if it is said that information is to be reviewed by a machine but is reviewed by a human.

She emphasized that reviewers must take responsibility for information leakage even though it is legal to use human review when a machine can’t tell. Users can report to regulatory authorities or utilize legal means to defend their rights if an Internet service provider violates the Personal Information Protection Law.

What can users do to protect their privacy? Before uploading their data to cloud storage, Peng advises them to take technical precautions to secure them. The simplest method, for instance, is to create a package using compression software, encrypt it with encryption software, and then transmit it using ciphertext to prevent being opened by a person.