_China_Chatbot_08
What happens when Chinese media meets Chinese AI
Hello! Welcome to another issue of China Chatbot, where I take a look at the policy driving China’s international AI push, notice a change in how state media is reporting AI stories, and get China’s best diffusion model yet to cross Beijing’s red lines with ease.
But first, a mystery. This National Day holiday (also the PRC’s 75th anniversary), CCTV launched an AI-generated video honoring the nation’s flag. It’s a boilerplate message in a state-of-the-art package: the first man to raise the national flag in Tiananmen Square back in 1949 is magically transported from the past to a limbo-land where the Chinese flag of 2024 thanks him in a breathy female voice for his service. “The country couldn’t have grown in 75 years without people like you.”
The mystery is in the credits. CCTV lists Chinese home-grown LLMs like Vidu, HailuoAI, and Kling for the video. The thing is, CCTV has been generating and broadcasting dozens of AI videos far more sumptuous than this one for months, but not once has it ever listed the model behind them. They did so when their AI videos were being generated by their very own in-house model, but any credits disappeared once their model went back to the drawing board in February. Who’s been behind CCTV’s output since then?
My hunch (which could very well be wrong) is that either CCTV hired a foreign company for these videos, or the Chinese government has an AI model it’s been keeping from the public. The former is more likely. As we’ve just seen, they are more than happy to publicize when they use China-made models — indeed, publicizing them would highlight the quality of Chinese AI, that it’s good enough for national broadcasts. But I can’t think of any Chinese diffusion models in the public domain from March that were capable of generating videos at the level CCTV was.
Why would they consistently not credit the algorithm they’ve been using for the past eight months? This is either because they are saving face for having approached a geopolitical competitor, or (for whatever reason) the model has been deemed a state secret.
And with that, on with the show. Enjoy!
Alex Colville (Researcher, China Media Project)
_IN_OUR_FEEDS(3):
Strange Bedfellows in Xinjiang
On October 14, senior executives from international news outlets converged on Xinjiang for the 6th World Media Summit, an annual networking event hosted by the Chinese government. This year’s meeting was all about how AI should be used by the media.
Leaders from Reuters, AP, CNN, TASS (Russia’s state news agency) and Al Jazeera, among others, rubbed shoulders with head officials from key Chinese state media, including the People’s Daily, China Media Group, Xinhua, and China Daily. Multiple speakers at the summit urged the need to balance harnessing AI as a force for good with combatting AI-generated misinformation. Outside the summit, Xinhua demonstrated how it could use AI to translate bulletins from its English-language news anchors (whose short videos often appear on Western social media) into multiple languages. In advance of the summit, Xinhua’s president met with representatives from AP, AFP and Reuters in Beijing for their own smaller dialogue on AI in the media — part of a “cooperation mechanism” between the four news agencies started in December last year.
TL;DR: China is using AI as a front for its global influence and networking campaigns. Media falls well within the plan
Application, Application, Application
Chinese state media have been placing an emphasis on stories about AI achievements in the real world. On October 11, People’s Daily Online published a piece titled “‘AI+’ is a new engine for industrial transformation,” with a list of accomplished examples in the government’s policy to push AI into every industry in the country. Multiple outlets republished a Xinhua bulletin from October 13 quoting the Ministry of Industry and Information Technology as saying the number of registered AI users in China had reached 600 million.
On October 15, People’s Daily Online covered a new LLM that optimizes production in the oil and gas industry, while CCTV reported on how Chinese astronomers have used deep learning models to discover new stars. Xinhua has also been promoting a report it released for the World Media Summit on how media can be used in the era of AI, approaching friendly foreigners at the conference for comment on its “important guiding significance” for the industry’s new technological development.
TL;DR: Not so long ago just releasing an AI model was enough to earn coverage. Not anymore
Robin Li Does Damage Control
In an interview with Harvard Business Review on October 16, Robin Li, the CEO of Chinese search engine giant Baidu, offered predictions for the generative AI market. Li said an AI bubble was “inevitable” given how the novelty is wearing off, and that this will lead to disillusionment like when the internet first launched in the 1990s. He proposed this is “healthy” since it cuts out the market’s dead wood, leaving 1 percent of companies that would be of huge value. Meanwhile, he claimed AI hallucination (when an AI model produces inaccurate information, a problem experts say is currently impossible to fully resolve) had “pretty much been solved.” He also said Baidu was doing a complete overhaul of all its products to include AI, including its flagship search engine.
TL;DR: Robin Li wants Baidu to survive any coming burst and come out the other side a major AI platform. To do that, he has to downplay one of AI’s biggest flaws to potential investors
_EXPLAINER:Global AI Governance Initiative (全球人工智能治理倡议)
What’s that?
In a nutshell, it’s a campaign for China to win international influence and business by becoming the lead player in global AI development, cooperation, and governance.
So along the lines of something like the Belt and Road Initiative or the Global Development Initiative?
Yeah, previous initiatives have mostly had the same aim: providing development opportunities for countries mostly in the Global South to raise themselves up, in turn creating what the CCP calls “a community of common destiny for mankind” (人类命运共同体), essentially a demand for states to engage in what they term “win-win” cooperation and dialogue, but not interfere in each other’s internal affairs.
So this one does the same, but with AI?
Exactly. It’s no coincidence Xi Jinping launched the initiative at the Third Belt and Road Forum in October 2023 (pictured above), saying China was willing to engage in “healthy, orderly and safe development of AI.” The accompanying policy document stipulates countries work together to ensure AI boosts human welfare around the world, but also to shore up (China’s) AI safety concerns and ensure products “respect the sovereignty of other countries.”
So what’s the initiative doing?
Multiple projects from China in the past few months have cited the Initiative in their rationale. In September, Foreign Minister Wang Yi cited it when he launched the “AI Capacity Building and Inclusiveness Plan” (人工智能能力建设普惠计划) at the UN, saying China would create an “open source AI community” with other countries, build joint AI labs, and provide AI education and products. Beijing’s “AI Safety Governance Framework” (人工智能安全治理框架), which catalogs current AI risks and proposes a raft of safety standards, does so in the name of the Initiative. Xinhua cites it as the rationale for their report on how global media can use AI responsibly and fruitfully, released this week at the World Media Summit. The Initiative’s shadow was also present in China’s creation of the “China-BRICS AI Cooperation and Development Center” (中国－金砖国家人工智能发展与合作中心) this summer, and in offers by Chinese tech companies to African leaders for new AI products on the sidelines of the Forum for China-Africa Cooperation last month.
So Chinese companies are going to get some good deals out of this too?
Most likely, in the same way they did with the BRI. The Chairman of Chinese telecoms giant ZTE has said he hopes to invest in AI products for Africa, pointing to high demand from the continent’s young population. He said the firm would “jointly explore” AI with African governments and companies, in fields like logistics and surveillance.
I assume China would be the senior partner in any joint exploration?
Absolutely, given the extent of their development in AI thus far. Where they will take these projects remains to be seen.
_ONE_PROMPT_PROMPT:For the past fortnight or so I’ve been tinkering on a project that uses AI tools available on the Chinese internet, to see how easy/hard it would be to create a deepfake inside the Great Firewall. In this lightning-fast field, it will probably have the shelf life of iceberg lettuce, but hopefully it shows how low the bar has gotten, how easy it is to override China’s current guardrails, and where we are for realism. You can read all about it here.
My favorite moment was tricking AI start-up Minimax into making Donald Trump affirm Taiwanese statehood:
Still not quite there for realism, but it’s also exactly the sort of thing Beijing doesn’t want Chinese AI models to do. So still work to be done there.
(Source: Minimax’s 海螺AI)