AI shouldn’t be a black box: Infosys co-founder Nandan Nilekani is impressed with China’s DeepSeek, but...

Infosys co-founder Nandan Nilekani has criticised the trend of developing increasingly complex artificial intelligence (AI) models, arguing that this does not necessarily lead to better results. Speaking at the People+AI Mela in Bengaluru, he expressed concern over how many AI models remain closed systems, hidden from public scrutiny.

“They keep doing more of that, but it is not that it is going to get any better. They are all typically kept within the four walls of a company. It is like a black box for the rest of us,” Nilekani said, as reported by The Economic Times.

This lack of transparency, he warned, is limiting innovation and preventing broader contributions to AI development.

Learning from China’s Approach

While critiquing the secrecy surrounding many AI models, Nilekani praised China’s approach to AI development, particularly the DeepSeek project. He pointed out that China has managed to build effective AI models with fewer resources while maintaining transparency about their methods.

“They have done the world a service. So that goes back to commoditisation,” he stated.

According to Nilekani, AI models are increasingly becoming commodities, with hundreds of new ones emerging worldwide. The openness demonstrated by China’s DeepSeek project, which publicly shared its AI-building practices, has contributed to demystifying AI development.

“I have always believed that models are a commodity. This is proving to be true. Hundreds of models are getting built. China’s DeepSeek demystified model-building. Last week they tweeted all their open-source practices. The knowledge to build models is getting rapidly commoditised,” he added.

The Need for High-Quality Data in India

Beyond model-building, Nilekani emphasised the importance of high-quality datasets, especially in Indian languages, to support AI growth in India. He cited AI4Bharat, an initiative by IIT Madras focusing on open-source AI development for Indian languages, as a key step in this direction.

“The hard part is uses. How do we use this to deliver value for people? That is where we can lead. We can always replace the model,” he noted.

AI4Bharat has been backed by Nilekani for several years and will continue to receive his support for at least three more years. The initiative aims to enhance AI accessibility for diverse linguistic groups in India, addressing a major gap in AI research.

Startups and the Future of AI in India

Nilekani also touched upon the rapid growth of India’s entrepreneurial landscape. The number of startups in the country has surged from 1,000 in 2016 to over 150,000 in recent years, driven by advances in digital technology and increasing capital inflows.

“They will use mass-scale technologies that we have deployed in interesting use cases. There is also going to be a lot of capital coming in,” he remarked.

Despite the evolution of AI models, Nilekani believes the real challenge lies in their application. The focus, he insisted, should not be on simply building advanced models but on harnessing AI to address real-world problems and improve lives.