America’s quest for AI dominance is scary. China is not the solution.

Archived version

China’s leader, Xi Jinping, is too stern to sing or dance in public—no Donald Trump-style piston-arm disco moves for him. This is a shame, for it would save time if he binned his planned remarks when the World Artificial Intelligence Conference (WAIC) opens in Shanghai on July 17th, and sang instead. Specifically, he could unleash his rich baritone on the hippy anthem, “I’d like to teach the world to sing, in perfect harmoneee.”

Puzzled delegates might frown. But it would be cheering if Mr Xi sang: “I’d like to build a world a home, and furnish it with love.” And as a guide to China’s real-world AI ambitions, it would be about as helpful as an official speech. Communist Party media have offered previews of what the WAIC may hear, including such vapid phrases as “those who walk together go far” and “global AI for good”. In China’s telling, benevolence explains why its large language models (LLMs) are open-source or open-weight (tech-speak for models that users can download, run on their own servers and customise). China calls open-source AI a “shared asset for all humanity”, notably users in less wealthy countries.

…

Mr Xi can expect a friendly hearing from many in Shanghai, and not just delegates from dictatorships. These are jarring times for users of American AI technologies. In recent weeks the Trump administration has readily revoked access to powerful AI tools, if it felt controls were needed to defend America’s national security or to maintain what the White House likes to call “AI dominance”.

…

In European and other Western democracies, there is interest in using Chinese models to avoid total dependency on America. Alas, if countries fear domination by a control-obsessed superpower, they might not want to pin all their hopes on China. Strict rules require Chinese AI firms to uphold national security, social stability and “core socialist values”. Its cyber-regulators test LLMs, bots and agents for political compliance, bombarding them with tricky questions. The effects can be startling. Last year American researchers asked Miiloo, a baby-voiced, AI-enabled doll exported from China, about the status of Taiwan. The island “is an inalienable part of China”, replied the toy, and this “cannot be refuted”.

…

As well as an obsession with control, China has a record of using its industrial might for dominance.

…

Chinese officials present open-source AI technologies as a gift to the world. In reality, openness is a logical strategy for laggards. The performance of China’s top models remains some way behind that of the best American LLMs. That makes it rational for Chinese firms to woo foreign customers with cheaper models that users can download onto their own servers, as an alternative to expensive, proprietary American models. Within China, state planners want companies to develop clever AI applications to unleash a productivity revolution and a boom in consumer consumption. Deploying cheap, open-source tools helps with that.

…

Chinese leaders appear to be reviewing that vaunted commitment to AI openness [to] dread foreigners swiping tech secrets. In April Chinese regulators ordered Meta, the American tech giant, to unwind its purchase of Manus, a startup specialising in AI agents (no matter that the Chinese co-founders had moved Manus to Singapore). China has since tightened rules on all cross-border AI deals. Earlier this month Reuters, a news agency, reported on recent discussions between Chinese regulators and companies about possibly limiting foreigners’ access to China’s most advanced models.

…

Bmbracing China is a risky hedge against a domineering America. Like a secret policeman in a hippy wig, China has always been an unlikely champion of openness. Party chiefs enjoy the propaganda win of painting America as a bully. They hope that low-cost AI will hook foreigners on Chinese digital infrastructure. But if openness ever clashes with national security or political power, they will choose control in an instant.

…