HONG KONG, Feb 27 (Reuters Breakingviews) - The open-source software movement has been an unprecedented driver of global innovation and productivity growth. By sharing code that runs databases, smartphones and designs semiconductors for free – and letting anyone contribute to upgrading it and patching its bugs – OSS, as it is known, has spared companies from the need to reinvent countless wheels, accelerating the growth of the $475 billion global software industry. Yet rising geopolitical tensions are putting the movement’s future economic contribution at risk.
As with other parts of the supply chain, Chinese and American code bases are quietly but deeply intertwined. China is the second-largest contributor to GitHub, a shared repository which Microsoft (MSFT.O) bought in 2018 for $7.5 billion. Nearly all Chinese smartphone manufacturers use versions of Android, an OSS operating system which generated $22 billion in profit for Google from its launch in 2007 through early 2016, according to an estimate revealed by a lawyer for Oracle that year. Google's owner Alphabet (GOOGL.O) does not break out Android revenues generated by advertising and app downloads from its store.
Originally more takers than sharers, mainland tech giants like Alibaba (9988.HK), and Tencent (0700.HK) now contribute generously to OSS projects overseas; China Inc has also shared frameworks for AI development, including search engine Baidu’s (9888.HK) PaddlePaddle and image recognition specialist Megvii’s MegEngine.
There are two big issues. First is security. As Western democracies use open-source code to accelerate strategic industries like artificial intelligence and semiconductor chip design, they also gift it to Beijing at a moment when Western AI is being deployed on the frigid battlefields of Ukraine, and the People’s Republic may be mulling military aid to Russia.
The second problem is moral. Chinese developers participating in international OSS machine-learning projects often contribute “model weights”: mathematical formulations that neural networks learn from training on China’s massive and opaquely sourced domestic data reservoir. The most impressive AI models require billions of examples from which to learn model weights and require large upfront investments, so companies generally welcome Chinese contributions. But software industry insiders told Breakingviews they are worried that these weights could be derived from data collected from sources like government surveillance cameras, some of which are allegedly trained to watch for ethnic minorities by companies like Megvii.
The United States fired a warning shot at telecoms giant Huawei in 2019. As with most Chinese handset makers, founder Ren Zhengfei relied on Android’s open-source code to run his company’s smartphones. But President Donald Trump’s administration put the company on a blacklist that effectively prevented it from using the Google services that complemented Android, in particular its app store.
The move crippled Huawei’s overseas smartphone businesses and telegraphed to China’s tech sector that the American government was willing and able to weaponise OSS. The same year, GitHub began blocking developers in Iran, Syria and Crimea. Today OpenAI, the somewhat misleadingly named Microsoft-backed company behind the popular ChatGPT bot, won’t allow residents of China and Russia to create accounts to use the tool. Beijing has reciprocated, per a Nikkei report, by instructing its own tech companies not to incorporate ChatGPT into their platforms - a rare example of a mutual firewall.
Huawei has since built its own OSS operating system called Harmony, and GitHub now faces a Chinese rival startup called Gitee, which enjoys backing from the Ministry of Industry and Information Technology. On nearly every front, China is moving to duplicate OSS functions, including chatbots and building communities it can control. Chinese developers are already complaining about censorship on Gitee.
“The United States government is much more comfortable going after companies than software because [the latter] is a lot harder conceptually for U.S. agencies,” notes China risk consultant Isaac Stone Fish. Blocking Chinese access to OSS libraries would be legally complex, given the code has been donated to the global commons. It would be technologically tricky too given the pervasiveness of virtual private networks that conceal where a user is visiting from. Also, nothing stops a Chinese intelligence officer from downloading OSS resources while abroad. And yet given fraying relations, it seems inevitable that Western democracies will start making it harder for China to take free code; they might also simply produce less of it. Washington has already selectively blocked access to U.S. government websites in China.
Severing the intellectual link between the world’s two largest open-source communities, the United States and China, would be indirect but profound. One study in 2018 estimated that OSS adoption generated up to €95 billion of “positive impact” in the European Union and predicted that a 10% uptake in code contributions from the global tech community could generate an additional 0.6% of regional GDP.
Dividing the world into competing open-source camps would mark another reverse for free trade. It would also be a sad endorsement of the old adage that nothing is truly free.
Follow @petesweeneypro on Twitter
CONTEXT NEWS
Regulators have told major Chinese tech companies not to offer ChatGPT services, the Nikkei news service reported on Feb. 22 citing sources with direct knowledge, causing shares in Chinese companies building chatbots to fall. On Feb. 24 China's Ministry of Science and Technology said it saw the potential of ChatGPT-like tech and would be pushing for the integration of artificial intelligence into Chinese society and the economy.
Editing by Robyn Mak and Thomas Shum
Our Standards: The Thomson Reuters Trust Principles.
Opinions expressed are those of the author. They do not reflect the views of Reuters News, which, under the Trust Principles, is committed to integrity, independence, and freedom from bias.
Pete Sweeney