The launch of DeepSeek-R1 by Liang Wenfeng earlier this month, and later of Qwen 2.5 by Alibaba, has generated a debate on its effect and the state of AI development in India. The release of DeepSeek-R1 has already affected the financial markets for AI leaders, primarily technology giants of the USA.
The Indian financial sector was largely unaffected, thanks to blessing in disguise – we do not have a competitor to either of them, neither the Chinese AI companies, nor the US ones. We have been quite complacent in using US AI products without worrying too much about our own capability. The release of Chinese products has, thankfully, shaken us for good.
As with all things Chinese, it is difficult to assess all the claims transparently even though DeepSeek is open source. It is anybody’s guess as to whether the vulnerabilities have been left unaddressed willingly or incorporated to exploit them in future. Nevertheless, even if the claims are off by an order of magnitude, they are still worth analysis.
For records, as in circulation in the media, AI model development and training needed by DeepSeek-R1 required just USD 6 million as against hundreds of millions of USDs by American companies. Reducing data resolution, without compromising accuracy, reduced memory requirement to about one fourth. DeepSeek-R1 appears to be an agentic AI.
This means that DeepSeek-R1 is like an ensemble of specialists, with the ability to talk to each other in a language that the others will understand, cohorted in one place. This is starkly different from the western approach of creating a God like entity that knows everything.
It appears that the work has been in progress since 2015 when Liang’s company High-Flyer started creating hedge fund models based on mathematics and AI. Further on, if the reports are to be believed, he silently started accumulating Nvidia graphics processors in 2021 anticipating US restrictions that followed later.
The predecessors of the R1 version, like V2 and V3, have been in the market for some time, perhaps without earning much revenue. The whole sequence indicates that the generative AI development had been worked since 2016 or closely thereafter with close support of the Chinese government.
Rush for AI
The far-reaching implications of AI are already well documented. AI has the ability, and speed, of predicting, assessing, acting and counteracting. From agriculture to healthcare, from outer space to deep in the ocean and from past history to future, AI has its footprint everywhere. However, there is a security and privacy dimension to AI as well. Application specific AI development was launched by various government departments in India early enough.
Hopefully, some of the models may have been developed and may be in operation, although not much authenticated information is available in public domain. In order to take advantage of the agility of Startups and MSME ecosystem, Government of India announced a Deep Tech Fund of INR 1 lakh crore in 2024 budget. There are reports that DRDO has finalized at least 5 projects under this initiative.
The IndiaAI mission was also launched by the government with a fund of INR 10,300 crore in March 2024. IndiaAI mission has now called for proposals for Indic foundation models. Some private entrepreneurs have also launched their own initiatives. However, there have not yet been credible reports on breakthroughs in any Indian initiatives on generative AI.
Generative AI is the rush to dominate the world, financially, strategically, culturally and civilisationally. Hence, for a country seeking developed status and aspiring to be the 3rd largest economy, it is important to maintain its sovereignty in all spheres, from borders to financial to technology to culture and to civilization. Indian experts have long been divided on whether and how to take this challenge.
The time has come to take the challenge and win it early. The release of DeepSeek-R1 has shown that if there is a will, there are enough ways to materialize it within constraints. India has to show that it has will and capability both.
Risks
As already known, a large number of activities have already shifted to ChaptGPT and its siblings. The younger generation no longer makes minutes of meetings, drafts emails, carries out trend analysis, generates MoUs or other draft legal papers or even presentations. The first draft, and quite good at that, is generated by AI, which is later finalized by experts.
With the structure of DeepSeek-R1, it may not be far ahead that all these tasks are shifted to it. This brings the question of national security. All basic data related to courts, businesses, markets, locations, infrastructure etc. may now get relocated to Chinese servers, and Indian users may not hesitate in providing our consent for this without even reading, leave alone understanding, the terms and conditions of usage.
Fake narratives of sovereignty over disputed territories and disinformation may also have become a real possibility. Several media houses have already reported ambivalent answers of DeepSeek-R1 with regard to Ladakh and Arunachal Pradesh. When these answers will get altered by a mere revision of software code, especially during the times of conflict, is anybody’s guess.
Manipulation or control of several sectors like markets, public order, finance, energy, banking, infrastructure etc., by an adversarial regime may have become a real threat now with the demonstration of generative AI capability by China. Nexus of China and Pakistan, and now demonstrated hostility of regime in Bangladesh, may not be glossed over in this consideration.
Generative AI can be used effectively to change perceptions, mold public opinions and reduce cognitive abilities of Indian citizens and the government. Several commentators have already reported DeepSeek-R1’s vulnerabilities with respect to injection attacks, cyber threats, jail-breaking, guardrails, data poisoning, output manipulation, biased outputs etc.
Apart from security aspects, the Indian IT industry will have to change gears. We may no longer remain numero uno as IT service providers of the world. Our engineers and scientists will have to be retrained to remain relevant and best in the age of fast evolving AI.
What should we do
The technology will continue to grow and will continue to break boundaries of nations. The best defence is to embrace technology whole-heartedly. The Government of India may like to assess the effect of DeepSeek-R1, and other similar Chinese products, on national security and may take a decision with respect to it’s availability to the general masses.
If banning is not a possibility, the government may like to place filters at appropriate attributes to minimize security risks. Alternatively, a counter-AI AI system may be developed at breakneck speed to minimize or neutralize the harmful effects of DeepSeek-R1 and other similar AI tools, whether from China or elsewhere.
Despite the fact that there are several risks with DeepSeek-R1, it also makes sense to learn from the successful model of R&D demonstrated by China.
The government, perhaps, has taken the first step by announcing an AI mission and deep tech fund. The second step, freeing these initiatives from financial bureaucracy is more important. Hopefully, the government will accord priority to this requirement. Thirdly, a national vision on the Indic foundation model and all function-specific requirements, applications and possibilities should quickly be in place and known to all stakeholders including industry and academia. This vision should cater for current requirements but be flexible enough to cater for future.
Fourthly, quantifiable and measurable targets with authority to make quick decisions and absorb genuine mistakes and failures may be operationalized. Fifthly, the private sector, specially the MSMEs and Startups, may be seamlessly integrated to leverage their agility and innovation capability. Sixthly, the government may facilitate data access in required areas and to the extent required with safeguards as needed. Seventhly, unshackle the tech industry from bureaucracy. Let the decisions be based on science and technology and be taken by those who know them.
In technologies like AI, the speed is of essence. The quicker we master, the better will be our disposition. Can a nation of 1.45 billion afford to be laggards? Can a nation that once dominated the world in knowledge and wisdom remain oblivious to current requirements? Do we not have intellect to have our own AI models? Do we not have the will to generate our own solutions? Did it happen in the government system anywhere in the world? The answer to each one of them is emphatic NO.
The Government of India has already put the supporting structure in place. Hopefully, it will also take further measures as suggested to enhance ease of doing business in the area of AI. It is now for the labs, industry and academia to put brains and efforts together and show our capabilities to the world.
(The author is former DG, DRDO and Professor of Practice at IIT Delhi; Views expressed are personal)
