Artificial intelligence (AI) holds immense potential to transform India, offering solutions to challenges in governance, healthcare, agriculture, financial services, and much more. However, to reap AI’s full potential requires developing AI systems uniquely suited to India’s linguistic, cultural, and infrastructural realities. This calls for localized AI models, robust data sovereignty, and a distributed approach to computing infrastructure.
The world’s most popular AI models today are mainly trained using data in the English language. This means that the outcome of using these models is more skewed towards an Anglo-centric lens from a linguistic and cultural point of view. India’s linguistic diversity, with 22 official languages and hundreds if not thousands of dialects, makes localized language models, both large and small, essential for inclusive AI adoption.
On the one hand, it is critical to have large language models (LLMs) trained in Indian language data, tailored to understanding Indian nuances, and generating text in Indian languages. This kind of localized AI model reduces cultural biases and improves relevance in applications like government services and education. They ensure the outcome of these AI solutions are more relevant and accessible to all Indians, including those who do not speak English. For e.g. the Indus LLM by Tech Mahindra – an indigenous foundational model designed to converse in a multitude of Indic languages and dialects implemented using an innovative ‘GenAI-in-a-box’ framework.
On the other hand, small language models (SLMs) that are based on lower number of parameters, are more easily fine-tuned for localized applications like regional language translation and industry-specific customer service chatbots. It also requires significantly less computing power, making them ideal for our environment, using AI PCs, mobile or edge computing devices to enable AI at scale, enabling benefits across the society, just like the impact Aadhaar and UPI have made.
Developing these models requires the collection of vast datasets in Indian languages, achievable through crowdsourcing and collaboration between academia, industry, and government. Beyond the models, we should also focus on the applications of these models to solve real challenges at scale, that may be the tipping point for us.
India contributes approximately 20 per cent of global data, yet much of it is stored in overseas servers. Data is the most valuable input for AI, and we have an abundance of it. While the importance of data localization for security, compliance, and governance has been widely discussed, a lesser-known fact is that data localization also offers significant benefits for AI deployment.
For one, when data is stored and processed within the country, it reduces latency associated with cross-border data transfers, resulting in more efficient AI operations. For another, relying on data centres within India can be more cost-effective compared to using overseas cloud services in the long run, as subscription costs and transferring large volumes of data across international borders are high. Also having local data means local context is better understood and the applications we develop will be more relevant.
While initial capital expenditure for setting up local data centres can be significant, the Indian government and private sector are investing heavily in expanding data centre capacity, which will eventually drive down costs through economies of scale as more facilities come online and competition increases among local providers.
What is more, by relying on domestic data centres, organizations contribute to the local economy, creating jobs and fostering technological advancements within the country. This can lead to a more robust ecosystem for AI development, further driving down costs through increased collaboration and innovation.
Investing in local data centres — akin to building highways or other infrastructure, is critical to India’s future as the nation aims to be an AI powerhouse.
While expanding data centres in India is necessary, the future of AI lies in distributed computing—leveraging a blend of centralized facilities, AI PCs, and edge computing.
Here are the reasons. First, it can be expensive to run everything AI via data centres or the cloud. Whether it is owning, leasing local data centres, or relying on local cloud subscriptions; the operations and complexity can be out of reach for many organizations. Second, while it is already an improvement to rely on domestic data centres instead of overseas ones, sending data back and forth between where the data is generated and the data centres can still slow things down, and not ideal for time-sensitive applications like payment systems and healthcare monitoring. Third, even with data centres within India, some organizations would still rather have their data stored within their own premises due to security and privacy reasons. Beyond cost, latency and security concerns. there are also environmental challenges (for example, India has just about 4 per cent of the world’s water resources) of building large-scale central models of data centres. We have to reimagine the way data centres are built, and a distributed model of computing can be the solution for us.
Computing needs to be spread across locations and devices, using different computing resources for different AI use cases. With millions of PCs and edge devices, ~ 950K cell phone towers etc, every compute point can be an AI engine. As computing becomes more powerful, why train a smaller language model or run an AI agent or application in the data centre if you could do it right on your PC or at the edge? Edge computing enables real-time processing, reduces network congestion, and can continue to process data even if internet connectivity is down – so why only rely on data centres?
AI is complex, with different computing requirements for different use cases. To ensure AI is accessible for all, we must reimagine infrastructure that makes the most sense to supply India’s insatiable demand for computing.
According to the IDC Asia/Pacific AI Maturity Study 2024, commissioned by Intel, India is currently at stage 2 of AI maturity, at the AI practitioner level. However, to climb the AI maturity ladder, India must take steps to address the areas mentioned above.
India’s strong government support, vast talent base, and rich data sources make it a country poised to reap tremendous benefits from AI. By focusing on these priorities, India can better take advantage of the incredible potential that AI can bring, and use it to serve its citizens more equitably.
India is quickly expanding its AI infrastructure at the moment, and it is more important than ever to ensure that the foundation laid now will serve the people of its nation long-term, and that it is accessible, sustainable, and inclusive.
The author is the Vice President and Managing Director, Intel, India Region. Views expressed are personal.
Link to article –
