Learning from DeepSeek

Illusions & Delusions

The high-performance AI model from Deep Seek—a recently set up AI Lab in Hangzhou, China—has hit global headlines. Its founder used to run a hedge fund but re-purposed himself post-Covid and last week, released a low-cost AI model, seemingly faster than the US versions while using lower computing power, much unlike the multi-billion-dollar US models. Interestingly, other Chinese companies have also announced their AI models. Even if all these are creations of reverse engineering, the so-far unchallenged monopoly of US tech majors has been broken. While this may or may not prove to be another ‘Sputnik moment’ for the US, we, in India, need to ponder over the implications. We also need to be careful about the lessons we draw.

Creating our own AI models will not be difficult if Indian outfits/startups get down to it in project mode, but this in itself would be a red herring. As would questions like “Why are our IT majors not in the AI race?” All companies work on business models. Our IT companies historically focused on tech-use, not on tech-invent. We have, so far, been happy with our consumer-driven growth story and never entered tech-creation barring our atomic energy and space exploration programmes, where we have done exceedingly well and are globally respected. We, therefore, now need to introspect on whether we wish to promote a tech-invent culture in a variety of sectors that can anticipate, not only respond, to technological change. Creating an innovation-friendly economy sounds nice, but it needs a mind-shift.

This could begin by understanding why our national expenditure on R&D (0.7% of GDP) is significantly lower than the competition, despite a variety of tax concessions and programmes that include Make in India, Start-up India, Faster Adoption and Manufacture of (Hybrid and) Electric Vehicles (FAME), and the Mission on Nano Science and Technology. The private sector is also reasonably active, with 4,000-plus R&D establishments. Individual companies demonstrate varied performance levels. Several pharma, automobile, and IT majors spend significant amounts, often up to 8% of sales on R&D, and compare well with their global peers; India is perceived as globally competitive in these three sectors.

We need to recognise that the actual cause for our low R&D lies in the very structure of our manufacturing sector. About 90% of our companies are MSMEs, of which 90% are micro and small. This structure was created as it has so far been felt that promoting labour-intensive manufacturing is the most suitable national strategy to secure development. Other countries, unlike us, debate the benefits of low-tech versus high-tech manufacturing. Interestingly, most high-tech enterprises, by necessity, are usually promoted by technocrats, but also have high capital costs.

Our definition of what constitutes MSME is not globally equivalent though we adhere to global trading rules. They are less than half of US/EU rates though virtually the entire machinery is imported. The small sizes of our industrial units create another problem since small companies need more handholding. We have over two lakh industrial units housed in about 2,000 or more industrial estates/parks/SEZs; only a few are close to top-tier cities. So, companies located outside industrial estates may not be getting adequate access to technological guidance, unlike their peers in other countries. Thus, our globally competitive industrial spaces are relatively tiny compared to those in countries with a higher number of ‘likeable cities’ in relation to population. This is why component or tech manufacturing has not taken off despite a sustained policy push.

Given this scenario, policy modifications are worth considering if we wish to shift to a tech-invent mode. We could have two MSME definitions: the existing one for low-tech sectors, and an EU/USA equivalent for high-tech sectors. All current priority-sector privileges permitted to MSMEs today could apply to both. Further, we need to firm up a ‘locational policy’. We could identify a manageably large number of industrial estates suitable for technology manufacturing and stock them up with a higher quantum of a variety of public goods to facilitate efficient high-tech production. We could also follow the best practices of other peer countries and start creating a sufficient number of attractive living spaces that can double up as innovation hubs. Hangzhou is far from Beijing, Shanghai, Shenzhen etc but could still produce a giant-killer.

Finally, we could also review the goals of our educational institutions. These were as good as the global best but have started getting perceived more as cost centres and thus subjected to ‘massification’ to reduce the fiscal draft, instead of being continuously upgraded to rank higher than competitor countries. Resultantly, our laboratory-student ratios have worsened and our class sizes now run into triple digits, diluting quality standards, while fewer academic jobs are on offer. If we can reverse these disadvantages, our ever-increasing brain drain may well start reversing and our DeepSeeks could start surprising the world.