India Policy Watch #1: The Looming Spectre Of AIInsights on current policy issues in India—RSJOver the past couple of months, I have spent some time on the state of play of AI in India. This has meant doing a few things: sitting through real examples (use cases) of AI being used by businesses in India, understanding how organisations are adopting LLMs at scale through in-house platforms, seeing first hand some of the Indian IT services firms readiness to ride this wave, the kind of work Indian start-ups are doing on foundational technologies in this space and the funding environment. The whole experience was sobering. India will be a tremendous market for the consumption of AI applications and services, but it will struggle to be a supplier of innovation and intellectual property. On the balance, we will derive secondary and tertiary economic value from this innovation, which will be undercut by the loss of value we will incur because of the disruptive nature of this innovation on jobs and income. The primary value will be cornered by the US and China. Unless, of course, things change dramatically for us. I was surprised by the examples of AI at work in large, traditional organisations and how fast it has moved in the past 3-4 months. Most people confessed that even six months back, their view was that we were going through the standard technology hype cycle, and it would make sense to wait for it to ebb before figuring out what investments were needed in this space. But the advances in the past six months have changed their minds. Each recent update of different LLMs is a step change in capability. Two things stood out to me. One, these were traditional organisations operating in core sectors of the economy and not start-ups or companies with tech-driven business models. Usually, these would have been the last to join the party due to legacy drag, technical debt, or regulatory restrictions. But the almost ‘plug and play’ ease of use of these AI tools has overcome those barriers. Sure, the use cases are still one-off areas of application of AI and resulting automation, and it will take some time to have end-to-end processes automated through agentic AI. Yet, when you see a nifty AI tool review four different and somewhat complex documents across multiple internal systems and write up a report on a customer and its business that incorporates other available information outside of the organisation too, in under a minute, you start wondering how long will that department and its staff that supports the frontline sales team with such reports will last. Because the report generated by the AI tool is more cogent than human-generated reports of the past, it has wider sources of information for its analysis and is better organised in its layout and readability than what the staff has been doing so far. Or, the other case where the LLM is able to create good quality marketing collaterals that adhere to the company’s branding guidelines, with the most basic of prompts. The text is better than what a mid-rung ad agency would have come up with, and the life-like and relevant pictures are generated by the tool itself (indemnified by the tool too for any copyright violations). And all of it done within an hour instead of the three weeks of back and forth that it took between the company and the agency. To be sure, the output won’t win you awards at Cannes, but then that’s not what the marketing team is looking for, either. The range of use cases were spread similarly wide across multiple departments and the impact of the automation was already being felt in cutting down expenses in using a vendor or hiring someone to do such stuff. The CEO of a large NBFC that’s always led the way on technology adoption indicated to me that he believes his contact centre will have a quarter of its current staff strength in two years’ time because of what they are seeing agentic AI do already. This thing was no longer cutting-edge tech understood by few and applied by fewer. This was on the cusp of going mainstream. Two, most of the heavy lifting needed to get going in this space like, choosing the LLMs, creating an underlying platform and workflows that allow employees to access multiple LLMs through this in a multi-tenancy model and the proliferation of use cases that I saw, was all done by in-house teams with almost no support from an external consultant or partner. This is different from many of the past instances of technology innovations, where the complexity of implementing them or the depth of technology understanding needed to figure out what they could do for your business meant organisations looked for external help. The usual approach was for the CTO to farm out work to a few vendors who came in and showed what they had done for other clients. The innovation would then be restricted to a few in the tech team who understood the technology, or would translate to one large system implementation, which no one was sure about. That, in itself, was a barrier to adoption. The general sense I get now is that most organisations believe AI is a fundamental capability for the future, and it has to be built in-house. That apart, the ease of use is so apparent that the ability to translate a business problem into a technology-driven solution now requires almost no programming knowledge. I might be exaggerating things for effect here. But we are almost at the point where every employee can program at their desk for their needs. The interesting thing here was the number of cases where the in-house IT teams accomplished tasks in a matter of hours that would have otherwise been outsourced to an IT services or testing vendor. I saw at least two real cases where top Indian IT services providers were working on a project for a few months in the conventional manner, till the in-house tech team with coding assistants powered by LLMs intervened to deliver the same output in a few days. To stretch a point here, conventional application development and services (ADM), the bread and butter of the Indian IT services sector, was shrinking to a tenth of its size right in front of my eyes. I suspect there is one of two things at play here, which is stopping the Indian IT services sector from doing this on their own for their clients so far. It is likely that the Indian IT services vendors are holding on to their usual FTE (or headcount) based pricing model in the hope that large customer organisations will never reach a level of AI sophistication where the customer will use the tools to automate and show it could be done to them. This won’t last, given the relative ease with which adoption of these tools is happening. Another possible scenario is that the level of understanding of what AI can do for their customers is quite limited among junior ranks in Indian IT services companies. They don’t have the subject matter or domain knowledge like that of their clients, nor have their employers given them free access to multiple AI platforms to play around with. I suspect the truth for an Indian IT service player is a mix of all of these. The bottom line is that the Indian IT and BPM services sectors are in for a rude shock over the next three years. Quite easily, a third of the jobs already there and maybe half of the jobs that could have potentially been generated will be made redundant in this space. I don’t think the scenario will be too different for mid- and back-office roles across other sectors of the economy. If the economy was already dealing with a “jobless” growth scenario in the past five years, we are in for a nastier surprise as the full impact of AI on jobs across the economy plays out. The trend across most large employers (BFSI, IT, and Services) in the past few years has been a net reduction in headcount, as companies used the COVID-19 shock to tighten their belts and continued to do so. The relatively weak revenue growth across these sectors, with record-high profitability in recent years, has been a result of this. This has meant underlying real wages and job growth have remained sluggish, which is now showing up in retail consumption growth in the GDP figures. The AI impact, if it plays out as above, will make it worse. Larger companies will continue to make more profits as they get more efficient because of AI but they will eventually hit the truth of the real economy that’s struggling. I don’t think there has been any real analysis of the impact of AI on the Indian economy. We will be net consumers of AI by a margin. And that will look good in the short term with productivity gains across sectors and possibly, strong boosts in access to healthcare and education. But it will hurt in the medium run for an economy that needs to create jobs in millions. It is easy to be seduced by various global leaders visiting India and hyping up the possibilities of AI or sweet-talking India’s approach to AI by focusing on its ‘practical’ applications. These tend to be those who look to India as a market for AI tools and see it as a consumer. The prospect of India catching up on foundational AI capabilities and participating in primary value creation, however, looks weak at the moment. Here’s from the Mint this week (“India lacks innovation, spending to realise AI dream”) on how India compares to the U.S. and China on this:
The government of India’s AI mission, with its $1.2 bn earmarked for startups and the plan to empanel over 18,000 high-end GPUs to power AI research, is a good beginning. There have been similar missions in the past (Digital Bharat, Startup India, etc.), so I remain sceptical about the efficacy of such piecemeal efforts. Beyond it, we need a total ecosystem to support primary AI research, including access to the most advanced GPUs and accelerators at the rack level to train AI models, investments in fundamental research and filing of patents, building vast and high-quality datasets, expanding semiconductor manufacturing capabilities and focus on training next-generation AI talent. This will take time, but the investment will be worth the effort and this is one area where the government leading with investment will draw in other investors. India might not be able to match the US and China in becoming an AI powerhouse, but it can still be a meaningful player that punches above its economic weight. At this moment, though, this appears to be a bridge too far. Addendum- Pranay KotasthaneWe have a rare disagreement on this issue. Three points.
What is your view? Do comment. India Policy Watch #2: Getting Things RightInsights on current policy issues in India—Pranay KotasthaneAs the investigation into the tragic AI 171 crash proceeds, the regulatory agency, DGCA, has belatedly sprung into action. In a classic CYA operation, it has now warned Air India over repeated lapses in crew rostering and has ordered the removal of three officials. It has also issued a show-cause notice to the same airline during a spot check, which revealed that pilots on two international flights had exceeded their stipulated flight duty time of ten hours. It’s pretty obvious what’s happening here. The blame will be pinned entirely on the airline; some of its people might lose their jobs, and some others might even use this tragedy to question Air India’s privatisation. The DGCA will be hyperactive for a few weeks, conducting spot checks and issuing notices at a frenzied pace, after which we will be back to where we were before the crash. But what’s needed is a long, hard look at the air safety regulatory structure. To be sure, India’s air safety record is heads and shoulders above that of any other means of transport in the country, and some credit for that should indeed go to the regulators—they have done a great job within the constraints they face. Nevertheless, the airline sector needs independent, empowered, well-staffed regulators because it’s an inherently oligopolistic sector with influential players. Is the Indian regulator geared to govern this sector? KP Krishnan’s latest Business Standard column points out the institutional setup problems that need to be addressed if India wishes to become a global aviation hub in the future. This column is a masterclass in regulatory theory and should be read by all public policy enthusiasts. He writes:
These structural problems call for an overhaul based on the Banishing Bureaucracy framework by Osborne and Plastrik. At its core, it means decoupling the many roles a government organisation does into multiple purpose-focused organisations. Generally, there are four types of government organisations, depending on their primary function as shown below:
Policy and regulatory organisations are known as steering functions because they determine the direction of the sector in which they operate. For example, SEBI decides the rules that guide financial markets. Service and compliance organisations are known as rowing functions because they are concerned with execution to reach a destination that steering organisations have decided on. For example, BSNL is a service delivery organisation which helps the government meet the goals that the telecom ministry has set. Now, most government organisations in India cannot be neatly classified in these four categories. They sometimes perform all four functions at once (think RBI). So, a key idea for government organisation reform is to uncouple steering functions from rowing ones. DGCA faces similar problems. It is tightly connected with the civil aviation ministry and is responsible for multiple functions such as “regulation of air transport services to/from/within India and for enforcement of civil air regulations, air safety and airworthiness standards.” K P Krishnan’s article further points out how other countries have effected this uncoupling:
To regain credibility after the crash, the Indian government would do well to redesign its air safety regulators based on these principles. While the investigations will hopefully reveal the real cause of this crash, a modern air safety regulator will hold India in good stead as its economy and air traffic grow towards the Viksit Bharat 2047 target. Global Policy Watch: The Iranian Bomb That Wasn’tGlobal policy issues relevant to India—Pranay KotasthaneIsraeli attacks on Iran began on 13th June, in response to Iran's enrichment of weapon-grade uranium. Not only did Israel operate freely over Iranian skies, but it also effected a series of high-profile assassinations of military personnel using drones from deep within the Iranian territory. After the initial shock, Iran recovered and has launched a series of counterstrikes on Israel. The war has escalated, and a lot depends on the American stance. Honouring Trump’s golden words, “I may do it, I may not do it. I mean, nobody knows what I'm going to do”, I will leave this question for another edition. The current moment, though, got me thinking about the escalation dynamics between nuclear weapon states. Here are three observations from an Indian lens. One, the Iranian regime is cruel, but its bigger problem is incompetence. Iran was right to have a nuclear weapons programme once it became clear that Israel had one. It is a normal state of affairs—one country acquires nuclear capabilities and its primary adversary follows. This dyadic relationship is unstable when one nation-state has nukes while the other doesn’t. But once both Iran and Israel had demonstrated nuclear weapon capabilities, a new balance would have been established. Large-scale wars between them would have become less likely because I do not buy the argument that the Iranian regime is irrational and would distribute its nukes to its proxies. No major wars have happened between two nuclear-armed states thus far. So I wouldn't be surprised if Iran were making weapon-grade uranium this time. Instead, what’s surprising is that Iran couldn’t build one over all these decades. Fifteen years ago, it was caught unaware when the Stuxnet malware destroyed over 1000 centrifuges at Natanz. And now, it couldn’t keep its enrichment secret enough before its adversaries found out. It is well-known that the transition period when a country is developing nuclear weapons is a dangerous phase because it remains most vulnerable to preventive strikes by the nuclear-armed adversary at this time. Iran stayed in this trishanku state for too long. Either it should have called off its programme entirely or begged, borrowed, and stolen the necessary tech and materials from Russia, China, or even Pakistan. It’s now paying a high price for indecision and incompetence. Two, Iran’s predicament highlights the pivotal role India’s 1998 Pokhran tests played. All nuclear weapon states hate other contenders, and as the ongoing war shows, will go to any extent to prevent them from getting there. Had India not done the tests then, a growing China could well have done to India what Israel is trying to do with Iran. That decisive step did heighten tensions between India and Pakistan (which also went nuclear subsequently), and the subsequent economic sanctions caused some real pain. But soon enough, others had no option but to contend with the new reality of India as a nuclear-capable state. The George W Bush government was favourable to India, and played a significant role in making India’s voice heard on this issue. Within a decade of the nuclear tests, the civil nuclear deal with the US was on the table. Three, it’s helpful to compare the India-Pakistan equation with the Iran-Israel dyad. As Operation Sindoor demonstrated, despite the high level of tensions, the two nuclear-armed powers were keen to de-escalate pretty early into the conflict. Compare that with the Iran-Israel conflict, where more than 400 Iranians and 24 Israelis have been killed thus far, with no end to the war in sight. Nuclear deterrence still holds between two nuclear powers. The other difference is that both Pakistan and Iran used terrorists and proxies against their adversaries. But the nuclear overhang prevents a full-scale retaliation on Pakistan, while Iran had no such insurance policy. By egging on proxies before going nuclear, it attracted undesirable attention to its weapons programme. From a realist lens, Iran failed not because it tried to go nuclear but because it didn’t try enough. And that’s the lesson that a lot of other countries will take away from this war. HomeWorkReading and listening recommendations on public policy matters
|
