Discover more from Interconnected
🏛️Washington is Using China to Destroy Open Source
It is undemocratic and will only backfire.
The word “China” has been used as the boogeyman for a lot of tech policy initiatives in Washington DC – some smart, some dumb. The dumbest one yet is the latest push to restrict and regulate open source – the one remaining norm of openness, transparency, and global collaboration – because, well, China.
Last week, congressional China hawks started pressuring the Biden administration to erect more restrictions on RISC-V, an open source chip design framework popular in both the US and China, in the name of preventing China from gaining chip design IPs from America. (See my many previous deep dives on RISC-V over the last three years.)
This week, The Atlantic reported that the Department of Commerce is considering possible ways to restrict the future releases of open source AI foundation models, like Meta’s LLaMA, using China again as the justification.
I won’t go into the obvious impracticalities of regulating access to something as globally pervasive, interdependent, and ingrained as open source, because that doesn’t need a full post to explain. It only needs three words: cannot be done. After all, 96% of all software is already open source. (I have also written many articles on open source in the last three years; this one on how open source technology supported global Covid response is a good one to start with.)
What’s perhaps more worthwhile to dissect is why is America shooting itself in the foot to restrict open source, which by its nature is open, transparent, and democratic – values that America ought to champion? And who ends up winning in this sad state of affairs? For something as ill-advised as regulating open source to get this much traction, there must be something worth gaining and winning.
Big Tech Incumbents: we have discussed the competitive dynamic before between big tech companies that are building proprietary AI models and ones that are releasing open AI models in “Meta Wants to Make AI Moatless”. In short, open AI models erode the moat of proprietary models and, by extension, their profitability.
Open source technologies have done this to proprietary ones in the past: Linux to Windows, the many (many) open source databases to Oracle, etc. As AI becomes the next frontier of a new technology platform shift (Andrej Karpathy of OpenAI calls large language models the new operating system), proprietary model makers are acutely aware of this threat.
Historically, it was hard to put up regulatory barriers against open source technologies using antitrust, copyright, or other legal frameworks, because the software is free and open and net-positive for consumers and societies at large. Fast forward to today, free and open technologie are still net-positive for consumers and societies, but what changed is our sentiment towards the meaning of national security and AI safety. By playing up the China threat and general fear and confusion around AGI, big tech finally found a wedge to lobby Washington to restrict open source, despite all the ironies that go with democratic institutions restricting democratic norms.
This naked attempt at “regulatory capture”, which investors like Bill Gurley have made articulate arguments against, is nothing new. But this time around, it is very well executed. Need evidence? The Information reported that OpenAI’s annual recurring revenue now stands at $1.3 billion. Last year, before ChatGPT’s launch, its revenue was a mere $28 million. This has got to be the steepest ARR hockey stick curve in tech history, and with that comes plenty of incentives to erect regulatory barriers against open models to protect this gushing spring of profit.
Charlatan DC lawmakers: a willing dance partner to big tech’s regulatory capture desires is a cohort of DC lawmakers and bureaucrats, who became overnight experts on China, technology, and national security. This cohort exists in both the congressional and executive branches. Being the loudest voice on the one thing everyone in Washington agree we should hate, i.e. China, is clearly a booster to an ambitious politician's public profile and future career advancement. The thinness of real knowledge or understanding is a mere inconvenience that few people bother to care about.
In the late 2000s, I worked briefly on the Congressional-Executive Commission on China, a commission dedicated to researching and investigating matters, like religious freedom, ethnic minority rights, and political prisoners in China – issues that drew a bright line between a democratic America and an authoritarian China. Back then, a “China Hand” dinner gathering could all fit in a single Capitol Hill row house – a small, not so popular group of people who really cared. Very few of the loud China voices you hear today were on that commission, some of them were not even in Congress, one of them was still in school.
In a way, it’s hard for me to totally trash the behaviors of ambitious charlatans, who just saw an opportunity and seized it. That's what DC attracts and breeds. You play the hand you are dealt. All I say is: well played.
China: ironically, any restriction on open source AI in the name of national security and competition against China is a net-plus for China. For one, it makes China’s AI capabilities look way better than reality. Being worthy of sanctions is strangely a validation in and of itself, though in this case an undeserved one.
China’s AI tech scene has cooled dramatically since it saw a similar hype to Silicon Valley’s early this year. On the startup side, the only “exit” was an acquihire of Guangnian Zhiwai, a high-profiled AI model company pitched as “China’s OpenAI.” It was founded by one of the cofounders of Meituan, only to be bought by Meituan for less money than it raised from VCs, after the founder suffered from mental health issues and went to the hospital. On the big tech side, the most notable progress is how quickly China’s large cloud vendors, like AliCloud and Baidu AI Cloud, made LLaMA 2 available as a cloud service to their users to stay competitive. On the academic side, the Beijing Academy of Artificial Intelligence (BAAI) did convene a successful conference where folks like Sam Altman and Yann LeCun all participated, a solid display of convening power which we wrote about. But the BAAI’s large language model, Aquila, is by no means advanced and only performs well in certain English-Chinese language benchmarks – the only use case that Chinese AI models have any advantage in.
Restricting open source AI models, with China as the justification and target, will just escalate bilateral hostility further and produce a welcoming domestic rallying cry, when the country’s economy is still sagging and the government is struggling to revive it. One consequence may be focusing China’s talent and resources to actually make AI advances – the opposite of any technology sanction’s objective. Need an example of a consequence from a poorly-conceived and poorly-executed policy from the Commerce Department? See Huawei Mate 60 Pro.
Commerce Secretary Gina Raimondo told Senators in a hearing last week that she needed “different tools” to enforce US technology controls on China, in light of Huawei’s chip advances. Is she asking for tools to restrict open source? That will be ill-advised and ill-fated indeed.
The losers in undermining open source may sound abstract, but the effects are concrete, profound, and long-lasting.
Without the type of open collaboration, open transparency, and open governance model that is foundational to how open source projects work, there will be less innovation and less dynamism among the best talents both in America and globally. This is truly sad and ironic, since many of the most ardent supporters of stricter sanctions on all things China espouse the spirit of “American Dynamism.”
Many scientists and academics, not just in areas related to technology, but also in other disciplines like law, have been lamenting the harm that the worsening geopolitical tension is doing to routine knowledge exchange and advancing new ideas.
The vast majority of programmers and developers everywhere, but especially the ones in China, have been embracing not just the source code, but the value system of open source for decades – a value system that is, for a lack of a better label, more “American” than “Chinese”. Instead of courting them as allies, their identity and talent would be further shunned, ostracized, and wasted, if dumb proposals like restricting open source become real policy.
I’m not suggesting there are no legitimate national security or competitive reasons for restricting China’s access to critical technologies or even banning Chinese apps in America. I’ve been a supporter of banning TikTok, if proven guilty, and have suggested policy frameworks and approaches to regulating apps like TikTok.
But using China as the boogeyman to restrict and possibly destroy open source, for profit and for career advancement, is a bridge too far.
Three years ago, I wrote an op-ed in Wired Magazine titled “The Future of American Industry Depends on Open Source Tech” with Jordan Schneider, where we argued that:
“[Open source] should not scare American policymakers, because the core values of open source—transparency, openness, and collaboration—play to America’s strengths.”
Our advocacy has clearly failed. America does not seem to recognize its own strengths, nor know how to amplify them, but is instead being swayed by fear and insecurity, and acting against its own interests and values.
This troubling trajectory to out China China in the name of competing with China at the expense of open source is undemocratic and will only backfire.