China sets AI rules that protect IP, people, the planet, and The Party
Brainboxes will need a license, but their makers may get to share digital public goods
Chinese authorities published the nation's rules governing generative AI on Thursday, including protections that aren't in place elsewhere in the world.
Some of the rules require operators of generative AI to ensure their services "adhere to the core values of socialism" and don't produce output that includes "incitement to subvert state power." AIs are also required to avoid inciting secession, undermining national unity and social stability, or promoting terrorism.
Generative AI services behind the Great Firewall are also not to promote prohibited content that provokes ethnic hatred and discrimination, violence, obscenity, or "false and harmful information." Those content-related rules don't deviate from an April 2023 draft.
But deeper in, there's a hint that China fancies digital public goods for generative AI. The doc calls for promotion of public training data resource platforms and collaborative sharing of model-making hardware to improve its utilization rates.
Authorities also want "orderly opening of public data classification, and [to] expand high-quality public training data resources."
Another requirement is for AI to be developed with known secure tools: the doc calls for chips, software, tools, computing power and data resources to be proven quantities.
- Intel woos China with nerfed Habana Gaudi 2 AI chips
- Should leaders place bets on new PCs or generative AI?
- China striving to be first source of artificial general intelligence, says think tank
- China to Meta: Flattery needed to get you into our VR market
AI operators must also respect the intellectual property rights of data used in models, secure consent of individuals before including personal information, and work to "improve the quality of training data, and enhance the authenticity, accuracy, objectivity, and diversity of training data."
As developers create algorithms, they're required to ensure they don't discriminate based on ethnicity, belief, country, region, gender, age, occupation, or health.
Operators are also required to secure licenses for their Ais under most circumstances.
AI deployed outside China has already run afoul of some of Beijing's requirements. Just last week OpenAI was sued by novelists and comedians for training on their works without permission. Facial recognition tools used by the UK's Metropolitan Police have displayed bias.
Hardly a week passes without one of China's tech giants unveiling further AI services. Last week Alibaba announced a text-to-image service, and Huawei discussed a third-gen weather prediction AI.
The new rules come into force on August 15. Chinese orgs tempted to cut corners and/or flout the rules have the very recent example of Beijing's massive fines imposed on Ant Group and Tencent as a reminder that straying from the rules will lead to pain – and possibly years of punishment. ®