Large language models, or LLMs, are the brains of ChatGPT, a chatbot that is employed by generative AI technologies. The "language" component is vital since models, no matter how big or little, cannot be created without a corpus of data. That's one of the reasons India hasn't produced any notable AI advances in the two years since ChatGPT launched, which started a fresh AI competition. However, in the background, the government, nonprofit organizations, Indian startups, and international IT behemoths are attempting to modify technology to meet national requirements. India's development in the upcoming century would be influenced by the rate and scope of their success. It will impart knowledge to other developing nations as well.

There are two big reasons for India to develop its own AI capabilities. First, as a rising power it is wary of depending on foreign technology. Second, it could be transformative for development. “The real value comes from how you apply these technologies to make a difference to people,” says Nandan Nilekani, a tech grandee.

Use the example of preparing dinner to gain a better understanding of India's AI opportunities as well as constraints. Data is the basic component of AI. India is effectively growing its own food in the absence of a well-stocked pantry. The Indian Institute of Technology, Chennai's AI4Bharat research lab has dispatched workers throughout the nation to manually gather voice recordings in 22 different languages. Google is operating in a like manner. Both contribute to Bhashini, an initiative by the government to develop an Indian language translation system.

Next, a model, or recipe, is used to mix, simmer, and season the data. Models can be very simple or quite complex, involving a large number of ingredients and intricate methods. There are a ton of recipes in ChatGPT and Google Gemini. But simpler ones might do for India's needs. Using open-source models as a basic sauce—like Meta's Llama—and then modifying the recipe or the methods to suit regional requirements is one concept. This is the path that Bangalore-based startup Sarvam AI is taking.

And last, using power effectively is necessary for cooking. Similar to how heat is necessary to transform materials into meals, specialist computer chips are necessary for artificial intelligence. The kind that are required to develop and operate complex AI models are costly and scarce worldwide. The government announced earlier this year that it will purchase 10,000 of them for 50 billion rupees ($600 million), enabling the provision of computing power at discounted rates. Additionally, Indian entrepreneurs are looking into alternative chip varieties that might be more appropriate for their needs.

So what will all of this work result in? Initially, chatbots will be the most noticeable items, just like in the West. The distinction is that these will be designed with immediate, useful applications in mind, mostly centered around translating and streamlining interactions between the public and the government. Additionally, Indians primarily utilize the internet for audiovisual content rather than text. Thus, unlike Western AI products, Indian AI will be voice-first or voice-only.

Consider filling out forms, which seems to be the national activity of India. Enabling individuals to respond orally to inquiries in their native tongue, which are then automatically entered into forms, would increase accessibility and cut out intermediaries. The process could be less painful if extensive compliance rules checklists were automated or if requirements interpretation bots helped with the process. "With UPI, an indigenous digital payment system, we can now declare that something in India is superior to anything in the rest of the world for the first time. However, the reality is that nothing else is superior, according to Sarvam co-founder Vivek Raghavan. AI "has the ability to flatten that, if everything became easier to do," in his opinion.

AI may also be beneficial in fields like healthcare and education. Less than half of Indian fifth-grade pupils could read at the year two level, according to a 2022 survey. The state of the healthcare system is also terrible. The learning crisis might be resolved by low-cost, widely available individualized tutoring. Systems that assist with diagnosis, interpret test results, or handle administrative tasks could free up physicians to see more patients. The judiciary's slow pace could be increased by automating certain procedural duties, which account for up to 50% of judges' work.

Read more.