Tuesday, 8 July 2025

Betting on homegrown AI

IEC Tech

article here

How IEC Standards are helping India to become an AI powerhouse and jump ahead of the queue.

At the start of the year, the Indian government gave its electronics and tech sector a mission to develop the country’s first large language model (LLM). The artificial intelligence (AI) model, expected to be ready as soon as this winter, is part of a broader vision for the country to become a global AI powerhouse by 2047.

Key planks of the programme include the IndiaAI Mission, which launched in 2024, promising a budget of over ₹ 10,000 crore (USD 1,5 billion) a year to develop India's AI ecosystem through public-private partnerships. This builds on an earlier National Strategy for AI published by India’s public policy think tank to guide R&D in the emerging technology across healthcare, agriculture, education, smart cities and infrastructure.

The overall aim is to build standalone or sovereign capability in a technology which is seen as transformational across many sectors. “Unlike in the past, AI in India is no longer confined to a privileged few or dominated by global tech giants,” it argues. “[The] government is empowering students, startups and innovators with world-class AI infrastructure, fostering a truly level playing field, paving the way for innovation and self-reliance in this critical sector.”

Core to the strategy is for LLMs to be developed and hosted in India, tailored to work for the specific social, cultural and business diversity of its 1,46 billion people, who converse in 22 officially recognized languages. This approach is supported by Sovereign AI, a concept which refers to a nation’s capacity to be self-reliant by using its own infrastructure, data, workforce and business networks. In India’s case, the objective is to be competitive with China and the USA, which are currently leading the AI race.

Growing adoption of AI in India

India’s AI market is growing at a rate of 25-35% per year and is projected to reach USD 17 billion by 2027, according to a trade association for software and services. In a separate report, the association says it expects the broader technology industry in India to deliver USD 300 billion in 2026, driven in part by steady advances in AI, as enterprises expand their tech initiatives at scale.

Gnani.ai is one of hundreds of AI-focussed start-ups in India. “People are excited about AI and impatient to adopt it,” says founder and CEO Ganesh Gopalan. “Every board of every organization is putting pressure on their employees to use more AI in the workplace. It could be for cost reasons, to increase revenue or to improve customer satisfaction. Whatever the end goals are, I see a lot of interest in using AI in India.”

Anecdotally, there appears to be a lot less fear in adopting the tech than in western economies, which, Gopalan suggests, is partly because of the country’s rapid growth. “The question of AI replacing jobs comes less into the equation in developing markets like China and India compared to developed markets. In India, the discussions are always around return on investment (ROI). We are a nation culturally obsessed with ROI and that’s not necessarily the case in developed markets.”

Handling multi-linguistic complexity

Gopalan believes that Indian sovereign LLMs are a huge opportunity. He says global models like GPT (developed by OpenAI), Claude (Anthropic), Gemini (Google) and Meta (Llama), trained on largely English language and Western cultural databases, work less effectively in a country as diverse as India. “The world is not just one in which English is spoken. There are complexities within languages. So, when you translate global LLMs into other languages, they often fail because there isn't enough training data. The same is true for countries in Latin America or Africa. Even the basic things that global LLMs do for the US or the UK just cannot be done for many markets around the world.”

His company develops AI models from an India-first perspective. He explains, “The number of tokens [base units of text that LLMs use to understand language] that you would use for a typical LLM is very different when it comes to India. An AI model in Tamil or Telugu local language requires very different inputs and parameters.”

Gnani.ai has launched a speech-to-speech LLM that it claims can handle over 30 million voice interactions daily and supports 14 languages for businesses across India and the US. Speech-to-speech (or voice-to-voice) AI technology enables humans to converse with digital chatbots and other AI systems. As a subset of its LLM, Gnani.ai has developed small language models (SLMs) targeted at specific verticals, including the banking industry. It is in the process of launching another for insurance, with a retail model in the works.

“Every industry has a different vocabulary, phraseology, pitch and tone so in the context of customer management, for example, each will be different,” Gopalan says. Beyond addressing a specific audience with local and sector specific languages, the development of India-first AI models is also deemed important to counteract potential cultural bias, or lack of nuance, in global LLMs.

Addressing bias, a key concern

“It is incredibly important to take bias into account,” Gopalan adds. “Even without getting into the ethics of copyright, the first and most important component in your model is bias. It is a huge issue since you can’t ever be sure of the exact weight [or bias] on which a global model has been trained. You don't even know if it's biased until you use it. You can make an assumption, but basically, it's an unknown factor.”

A practical use case his company is trying to solve is to automate the underwriting of loans or insurance at a bank. “What if your underwriting process actually gave people loans they could not repay or unfairly denied people access to finance because the AI model had a bias? It’s not only highly unethical on your part, but it’s going to negatively affect your business.”

He advocates training models for corporations using the data they own and control. “These SLMs are not trained on trillion-parameter models [like LLMs] but on a few billion-parameter models that the [corporation] owns to make sure that biases are under control. That's why the concept of Sovereign AI is important. You can’t just write an application programming interface (API) for a global LLM.”

Gopalan says private industry, government and education are in lock-step in developing the country’s AI capability. “Educational institutions will adapt their course curriculum to include more AI components to feed demand from the private sector for jobs,” he explains. “Since AI is fundamentally software and coding, things at which Indians tend to be broadly good at, the future is looking good.”

Indian investment in education for AI

The government is building homegrown expertise to fuel research, innovation and the workforce in concert with multinationals. Mumbai’s Universal AI University, founded in 2023, was the first in the country to integrate AI across its curriculum, with the backing of a leading Indian IT company. A Fortune 500 company funded Mumbai’s Jio Institute (named after the multinational’s telecom division), including the Centre of AI for All, which advances “India-centric AI capabilities”. The Thapar Institute of Engineering and Technology has its AI programmes supported by an American computer processing giant. Another multinational is backing a training scheme to boost the AI skills of over 2 million Indians, and one of its leading researchers is working with the Bill & Melinda Gates Foundation to build “gender-intentional” datasets in five Indian languages that together are spoken by over 1 billion people.

Standards for responsible AI

As Gopalan made clear, an AI model is only as good as the data it is trained on, and bias can creep in, creating all sorts of problems, depending on the application area. International benchmarks which set guidelines to check the data quality and avoid any form of bias are a way of ensuring systems work efficiently and ethically. The joint ISO/IEC committee, established to develop standards in the area of AI, has published ISO/IEC 5259-5, which provides a data quality governance framework for analytics and machine learning to enable organizations and companies to direct and oversee the implementation and operation of data quality.

To identify and prevent bias in outcomes, the committee has published ISO/IEC TS 12791, which provides mitigation techniques that can be applied throughout the AI system life cycle in order to address unwanted bias. This document is applicable to organizations of all types and sizes .

AI is making huge strides in India. By favouring homegrown systems to compete with worldwide tech giants, India is rapidly becoming an AI force to be reckoned with. Standards can help it avoid some of the mistakes made by companies based in the USA or China, which started earlier in the race but initially lacked internationally recognized benchmarks to guide them.

 


No comments:

Post a Comment