Why China's privacy-first approach to AI could be its biggest advantage

Debate around AI governance, regulation, and security is heating up as DeepSeek’s launch faces international bans and EU’s AI Act sparks backlash and praise. Most agree we need stronger laws around privacy in AI but could there also be an argument against stringent privacy restrictions?

Ravi Jagannathan, an enterprise-focused AI/ML/Analytics Lead, argues that China's approach to data privacy could be giving it a significant edge in AI development, particularly in data-intensive areas like healthcare.

Leapfrog: "China, because of its centralized government structure, has a distinct advantage," Jagannathan explains. He suggests that while various countries maintain leads in different aspects of AI technology, these advantages aren't insurmountable. "You can maintain a lead, but competitors can leapfrog. I would back China as more capable of doing this simply because they don't have the same data privacy restrictions. They can access whatever data they need to train their models."

Data bottleneck: According to Jagannathan, the primary hurdle in AI advancement isn't actually the technology itself but access to comprehensive training data. He points to healthcare as a prime example where this disparity becomes evident.

"The main bottleneck is not the technology but the data to train your models," he says. "In the US, you cannot access a single patient's complete data. Your eye images are with your ophthalmologist, your lab results are stored elsewhere, and your MRI scans are in yet another system. There's no single repository that contains the complete health information of an individual."

This fragmentation of data across multiple specialized healthcare providers makes it difficult to develop AI that can understand the interconnections between different aspects of health data, impacting the US’s overall ability to advance in research. 

"Look at any significant AI paper in medicine—it typically comes out of Singapore or China," he notes. He attributes this to the centralized healthcare systems in these countries. "The Singapore health system is highly orchestrated by the government, as is China's. Their regulatory frameworks make it much easier to provide data for AI training than in the US."

Western response: Western tech companies have attempted to address their data disadvantage by training models on publicly available information, though not without controversy. Jagannathan describes how major AI systems have ingested copyrighted material while implementing filters to technically comply with copyright laws.

"What US companies have done is utilize data without compensating the original creators, using it to train their AI models," Jagannathan notes, describing this as operating in "a gray area where they're following the letter of the law, but not necessarily its spirit."

Eroded moats: The emergence of DeepSeek is challenging the conventional wisdom that massive data and computational resources are necessary to compete in advanced AI. According to Jagannathan, established companies have relied on two main competitive advantages: "Their strategy has been to amass enormous amounts of data that others cannot access, creating a competitive moat. Then they build systems requiring massive computational resources—hundreds of thousands of GPUs—making it prohibitively expensive for others to compete."

However, DeepSeek's approach suggests these barriers may be eroding. "DeepSeek is challenging the notion that you need that much computational power," Jagannathan explains. "The shocking reality is that the protective moat these companies relied on is disappearing. That's why we're seeing such a strong reaction from financial markets."