#294 RecalibrateProduction-linked Incentive Schemes, Grok and Indian Sensitivities, India's Semiconductor Policy V2.0
Course Advertisement: Intake for the 9th cohort of Takshashila’s 48-week Post-graduate Programme in Public Policy (PGP) is open. It’s a fantastic pathway for those who wish to make a career switch to public policy. Check all details on the website.India Policy Watch #1: Good RiddanceInsights on current policy issues in India— Pranay KotasthaneThis investigation of the Production Linked Incentive (PLI) Scheme by Reuters caught my attention:
It adds that PLI winners had achieved merely 37 per cent of the production target while a mere 8 per cent of the total PLI allocations had been paid out thus far. The reasons cited are “excessive red tape and bureaucratic caution”. Long-time readers might recollect that we have several posts about the pitfalls of PLI schemes. In November 2020, we anticipated the three unintended consequences of this policy:
After four and a half years, we know these unintended consequences have played out exactly as we feared. As we have written before, it wasn’t wrong for the government to push for PLI to capture the ‘China+1’ narrative. Given that every other competing economy had brought out its own industrial policy, the PLI was India’s entry ticket to the manufacturing game in 2020. But the mistake was hoping an entry ticket would be enough to win the game, that PLI would compensate for India’s poor tax, trade, and doing business policies. Good riddance if the government finally decides to move on from the PLI. The only thing worse than a poor policy is extending that poor policy to hide its failures. The government is on the right track if it is indeed calling the PLI off. Nevertheless, it doesn’t seem that the PLI is being replaced by something better. The article highlights that the government is considering upfront capital support for the construction of the manufacturing facility rather than production reimbursement. The theory of change seems to be that such a move would allow companies to recover costs earlier. That would be a wrong diagnosis. The bureaucratic caution that slowed down PLI is a feature, not a bug. We can’t expect government officials to act like venture capitalists taking a punt on companies without due diligence and strict conditionalities. Neither should we expect Indian line ministry officials to transform themselves into DARPA project managers. Moreover, a corruption scandal in a flagship scheme will harm India’s manufacturing reputation more than slow reimbursements. Thus, shifting the goal post from production to plant construction will not affect the outcomes—the disbursements will remain way lower than the promised allocations. Instead, it would make sense if the government shifted its focus to deregulation. Identifying and removing the tax, trade, and policy barriers that harm doing business in India is the only sustainable way to increase India’s manufacturing strength. No more shortcuts, please. India Policy Watch #2: GrokkedInsights on current policy issues in India— RSJGrok 3, the artificial intelligence model of Elon Musk’s xAI, provided some quality entertainment to its Indian users this week. As the maximally truth-seeking AI (Musk’s description), Grok 3 has touted comparable performance to its competitors from Google and OpenAI. But what Musk has also delivered to all of us is an AI model with fewer guardrails on how it collates its contents and presents them to its users. Grok 3 is trained on the usual data sources from all over the web like its peers but goes a bit beyond in using content available on X (formerly Twitter). It is designed to give its users a wider range of opinions in its search for “maximal truth”. To top this, Musk has added unique preset personalities for the voice bot to converse with its users. While ChatGPT has benign options like “casual and friendly” or “professional and formal”, Grok 3 offered what we knew we wanted but didn’t know how or whom to ask. And so Musk delivered unto us avatars like “sexy”, “romantic”, and what is possibly his favourite, “unhinged”. A few weeks back, when Musk made one of his regular appearances on the Joe Rogan podcast, I got an early view of what the sexy, unhinged version looked like when Musk, not entirely hinged himself, got into a banter with it. Musk could’ve made a TV show of it as the two traded insults and colourful jokes until Rogan, no lily-livered liberal himself, moved on. In its press release last month, xAI called this out. “Grok is designed to have a bit of a rebellious streak, so please don’t use it if you hate humour!" Heh. That last line was perhaps unconsciously directed to its Indian user base. As Grok 3 gained some traction in India with its users posting their responses on X, something Grok readily enables, we had all sorts of political opinions sought from it. And Grok responded without inhibitions. For reasons not clear to me, the early users in India who prompt-engineered the queries seemed to have some kind of anti-right bias. So we got all sorts of positive responses about Rahul Gandhi, Congress and—a clear sign that Grok 3 was putting itself into future trouble—Nehru. That was entertaining enough till the “unhinged” version started responding to users back in the same coin as it was being spoken to. Grok 3 soon turned into a chatbot with the ideological slant of a Karan Thapar and the argot of Nawazuddin Siddiqui’s character from Gangs of Wasseypur. This, as you can imagine, was a recipe for trouble. And that’s how the rumours about an FIR filed against Grok started. Thankfully, there wasn’t an FIR, but there is surely scrutiny from the IT ministry. This scrutiny will only intensify further because in these early days of AI adoption, there’s a perception of these models being viewed as the ultimate source of truth. Two things will follow. One, there’s going to be a lot of instances where the IT or the home ministry will use Section 79(3)(b) whereby online intermediaries cannot be held liable for user-generated content unless they fail to act on a court order or a government directive. Now what’s a government directive is open to interpretation. While the Supreme Court in the Shreya Singhal case ruled that a mere request from the government cannot be seen as a directive, the problem is that such a section exists where the interpretation cannot be watertight. Two, having been caught off guard this time around, the right-wing ecosystem will work overtime to prompt engineer Grok and other platforms to “correct” any bias that it might have. Historical revisionism, now done by Bollywood films and WhatsApp warriors, will get automated and made redundant by these models. But what might be the more long-term impact of this week’s fun and games is how AI regulations will take shape in India. EU, which has taken a lead in AI regulations with the EU AI Act that was endorsed by its parliament last year, focused on two key tenets in its design: consumer safety and a risk-based assessment of AI apps. Most of the use cases will come under the low-risk category, including AI to help with recommendations, nudges, and controlling spam. These will be regulated in a light touch mode, and companies must follow a code of conduct and voluntary self-disclosure on how they use AI. On the other end, certain systems will be banned, including things dear to the EU—policing of citizens, social scoring of people based on their public behaviour and using facial recognition content from CCTV and the internet to build databases. Much of real-time facial recognition has been banned except for limited use by law enforcement with prior approvals and other controls. There is also a category called ‘high risk’ systems that aren’t banned but will be tightly regulated. These involve the usage of AI in critical infrastructure like utilities or already regulated areas like banking and healthcare. I suspect in India, the focus of regulations will not be on safety but rather on “freedom of speech” issues. I forecast there will be a Bharat layer that most AI apps will be forced to apply to ensure that the most fragile of things, Indian sensibilities, aren’t hurt. Some players will mildly oppose this AI with Indian characteristics, but they will eventually fall in line without making such a Bharat layer explicit to the public. The trade-off that will be offered for this curb on “freedom” will be a more laissez-faire approach to safety and monitoring of the underlying algorithms than the EU. This will suit most Big Tech players quite well, who remain keen to participate in the Indian market and are happy to pay the price of an entry ticket set by India. India Policy Watch #3: What Should Semiconductor Policy v2.0 Look Like?Insights on current policy issues in India— Pranay KotasthaneApparently, an updated version of India’s semiconductor policy is in the works. The MeitY secretary had this to say to Business Standard at the IESA Vision Summit held in Ahmedabad a few days ago:
Thus, asking what the next policy package should look like makes sense. Many industry watchers argue that the next logical step would be to extend the upfront capital support currently provided to fabs and assembly plants to adjacent stages like semiconductor manufacturing equipment and materials. The theory of change is that such support is essential for creating an ecosystem around the one fabrication unit in Dholera. That would be a short-sighted move. Here’s why. We can infer three aims from the original policy:
Of the three, the second aim makes no sense, as India is no longer in a foreign-exchange-starved situation. Buying phone displays and display driver chips from China is not a strategic vulnerability, as plenty of alternatives exist. It’s the other two aims that need the government’s attention. The next version of India’s semiconductor policy should do fewer things and do them well. Instead of spreading itself thin across the entire supply chain, the government would do well to focus on some segments. For example, there is no need for extraordinary fiscal support to sub-stages of the manufacturing supply chain (materials and manufacturing equipment). These stages should be left to the market. Budgetary support should only target segments with evident market failures or significant strategic vulnerabilities. Markets self-correct and drive innovation efficiently. Government intervention, therefore, must be surgical rather than sweeping. A likely side effect of promising aggressive capital support schemes might soon be that companies may start citing the absence of such support to delay their investments. While incentives are helpful, we must ensure they don’t become crutches for companies. The second focus should be capitalising on India’s chip design strength by upgrading the Design-linked Incentive scheme. Under this scheme, the government planned to nurture 100 domestic semiconductor design companies and help at least 20 companies scale to achieve turnovers exceeding ₹1500 crore within five years. However, after 3.5 years of implementation, only 17 companies have received provisional support, with actual disbursements lagging significantly. No disbursement occurred in FY23, and even in FY24, revised estimates suggest only 25 per cent of the allocated ₹200 crore budget would be utilized. The FY25 estimate, again, stands modestly at ₹200 crore. The reasons for tardiness include strict provisions that put firms raising substantial foreign money out of contention, confusion on the government’s right to the company’s intellectual property, and entrusting a government company, which is also a player in this domain, as the nodal regulator of this scheme. I have outlined many of these problems here. This policy needs overhaul. Third, getting the Tata-PSMC (Tata Group’s joint venture with Taiwan's semiconductor giant) fabrication plant up and running is vital. The project’s ₹91,000 crore price tag, of which taxpayers pay 70 per cent, raises legitimate questions around viability and return on investment (we’ve discussed them here). Now that the funds have been committed, policy should ensure that the plant gets technology upgrades after it comes online. Though this plant will service less than 10 per cent of India’s chip demand, getting it to manufacture chips is at least three years away. There’s a lot left to be done here. Fourth, government attention is only needed in segments with strategic vulnerabilities. That’s where a compound semiconductor fabrication plant can help. Such chips are used for specific defence and space domains. In recent times, they have also been used in electric vehicles. Government spending on strategic semiconductors makes sense. To achieve real success, India’s Semiconductor Mission must emphasise our inherent strengths, prioritise targeted, strategic support, and delineate where government intervention is required and where it isn't. Less is more. HomeWorkReading and listening recommendations on public policy matters
|
