Why it will be a mistake to ignore GenAI lessons coming from China

The Chinese borrowed liberally from research in the West. For India, that's a good path to follow, while learning from Chinese developments, without compromising security in any way

Silicon Valley entrepreneurs and startups often hog the limelight when it comes to technology discussions in India. This is why much of today’s commentary is concentrated on the battle for Generative AI (GenAI) supremacy — and by extension, the race for Artificial General Intelligence (AGI) — in the US, to the exclusion of what is happening in other countries. While the French startup Mistral has received some attention because of Microsoft’s investment of late, the developments in China are either ignored or receive only limited attention.

This is a mistake. There is plenty of evidence that the US government’s bid to deny advanced chips and other resources to Chinese companies has not been particularly successful in slowing China’s progress in AI and GenAI. Enough high-end Nvidia graphics processing units (GPUs) are making their way to China, either via smugglers or intermediaries in other countries who use paperwork to obscure the trail to the final buyers.

Then, recently, The Wall Street Journal reported that brokers are offering cloud services and computing power from high-end servers with Nvidia chips in overseas countries to Chinese tech companies. These allow Chinese AI firms to access the high-end computing infrastructure, with the US authorities being none the wiser.

Meanwhile, Chinese telecom and chip giant Huawei unveiled the Atlas supercluster, developed specially for GenAI developers. Huawei  claims the AI supercluster can match any Nvidia GPU-equipped hardware in training foundational AI models with over 1 trillion parameters. Finally, Nvidia itself is designing an AI chip for China to get around the current US laws.

In terms of the adoption of GenAI models by organisations, a recent survey by AI analytics developer SAS Institute and market researcher Coleman Parkes found that Chinese companies were well ahead of their US counterparts in experimenting with GenAI, though they lagged behind in full-scale GenAI implementation. This survey should be taken with a pinch of salt, given that these are early days, and the sample of companies, around 1,600, was not particularly big. A more important indicator of China’s rise in GenAI comes from the World Intellectual Property Organisation (WIPO) in Geneva, whose data showed that China had filed six times as many patents for AI-related inventions as the US. (India is number five in terms of patents filed in AI).

China’s AI and GenAI research and innovation have not only thwarted all US efforts to slow it down, but could well surpass the Western world in taking the lead in this technology. Over 50 companies in China have been reported to be developing large language models (LLMs) and other foundational models. These include big companies like Alibaba and Baidu, which have developed their LLMs from the ground up, as well as dozens of startups that have taken advantage of Open Source LLMs available freely and built upon them.

The big advantage that China has over many is that it has a flourishing AI ecosystem, which, while initially inspired by ideas from Silicon Valley giants, has now developed formidable capabilities. Apart from that, GenAI depends as much on the availability of huge amounts of data for training the models as it does on high-end chips. It is here that China has another advantage — it is by far the largest generator of digital data.

The third advantage China has lies in its power generation capabilities — both in thermal power as well as in clean, renewable energy. This is crucial, given the gargantuan energy appetite of GenAI models.

What are the implications of the Chinese GenAI research progress for India? While it might be tempting to suggest that India should woo (or at least loosen controls on) Chinese high-tech investment and visas for Chinese engineers, as it is reportedly considering in areas such as solar panels and automobiles, this might not be the right approach. AI research — and GenAI research in particular — still poses many potential dangers. Collaborating too closely with China could risk giving them access to sensitive data in India, or even allow them to hack into our critical systems. For now, it might make more sense for Indian regulators and researchers to keep a close eye on Chinese research without rushing to grant access before building in enough safety guardrails.

More importantly, as the government works out its plans for boosting GenAI research in the country, the Chinese approach could give ideas on how to speed up the development by wooing back the best Indian minds working in this area in the US and other countries, and building up hardware capabilities with long-term policy focus.

The Chinese borrowed liberally from research in the US and other Western countries to build up capability. For India, that is a good path to follow, while also learning from Chinese developments in the field, without compromising security in any way.

The writer is former editor of Business Today and Businessworld and founder of Prosaic View