Crystal Hu and Sheila Dang
(Reuters) – TikTok is working on a clone of its recommendation algorithm for its 170 million U.S. users, which could lead to a version that operates independently of its Chinese parent and is more palatable to U.S. lawmakers who want to ban it, according to sources with direct knowledge of the effort.
The work to split the source code, ordered by TikTok’s Chinese parent ByteDance late last year, comes ahead of a bill to force the sale of TikTok’s U.S. unit that began gaining momentum in Congress this year. The bill was signed into law in April.
The sources, who were granted anonymity because they were not authorized to speak publicly about the short-video sharing app, said that once the code is split, it could lay the groundwork for a sale of U.S. assets, although there are no indications of that at this time. plans to do this.
TikTok declined to comment. The company previously stated that it had no plans to sell assets in the United States and such a move was not possible.
TikTok and its Chinese parent company ByteDance filed a lawsuit in US federal court in May, trying to block a law that would require the app to be sold or banned by January 19. The U.S. Court of Appeals on Tuesday set an expedited timetable for reviewing the legal challenges. to the new law.
MILLIONS OF LINES OF CODE
Over the past few months, hundreds of ByteDance and TikTok engineers in the U.S. and China have been told to start sifting through millions of lines of code, analyzing the company’s algorithm that pairs users with videos of their liking. The engineers’ goal is to create a separate code base independent of the systems used by ByteDance’s Chinese version of TikTok, Douyin, while excluding any information related to Chinese users, two sources with direct knowledge of the project told Reuters.
The previously unreported plan provides a rare glimpse into what a technical separation of TikTok’s U.S. operations could look like and shows the lengths TikTok will go to to deal with the bipartisan political risk it faces. US President Biden and other supporters of the law argue that TikTok gives Beijing too much access to massive amounts of data, information that could be used to spy on or influence American TikTok users.
Reuters previously reported that the sale of an application with algorithms is extremely unlikely. In 2020, the Chinese government added content recommendation algorithms to its export control list, requiring the sale or sale of TikTok’s algorithm to undergo administrative licensing procedures.
According to the legal filing, the source code for TikTok’s recommendation system was originally developed by ByteDance engineers in China and has been adapted to work in TikTok’s various global markets, including the United States.
ByteDance attributes TikTok’s popularity to the effectiveness of its recommendation engine, which bases each user’s content on how they interact with the content they watch.
‘OPEN SOURCE’
The complexity of the task, which sources described to Reuters as a tedious “dirty job,” highlights the difficulty of separating the underlying code that links TikTok’s U.S. operations to its Chinese parent company. The work is expected to take more than a year, these sources said.
TikTok and ByteDance have vowed to challenge the US law in court on First Amendment grounds. However, engineers continue to work under orders to separate TikTok’s US recommendation engine from ByteDance’s wider network, the sources said.
A previous plan to isolate user data in the US, called Project Texas, failed to appease US regulators and lawmakers. The company is now seeking to step up its efforts to demonstrate the independence of its U.S. operations from its Chinese owner.
At one point, TikTok executives considered open-sourcing some of TikTok’s algorithms or making them available for others to access and modify to demonstrate technological transparency, the sources said.
Executives communicated plans and provided updates on the code-sharing project throughout the team, in internal planning documents and in their internal communications system called Lark, according to one source who attended the meeting and another source who viewed the communications.
Reuters could not independently verify the internal communications.
Compliance and legal issues related to determining which parts of the code can be transferred to TikTok are complicating the work, one source said. Each line of code needs to be reviewed to determine whether it can end up in a separate codebase, the sources added.
The goal is to create a new source code repository for the recommendation algorithm serving only TikTok in the US. Once completed, TikTok in the US will launch and maintain its recommendation algorithm independently of TikTok apps in other regions and its Chinese version, Douyin. The move will cut it off from its parent company’s strong engineering base in Beijing, sources said.
If TikTok completes the work of separating the recommendation engine from its Chinese counterpart, TikTok management recognizes the risk that TikTok in the US will not be able to deliver the same level of performance as the existing TikTok because it is heavily dependent on ByteDance’s engineers in China. to update and maintain the code base to maximize user engagement, the sources added.