AI turning out harder to implement at scale than thought: Nandan Nilekani

Artificial intelligence is “turning out to be far more difficult than we thought to actually make work at scale and useful to everyone,” Infosys founder Nandan Nilekani said on Friday. For the technology to truly work, the “cost per inference” should be as low as ₹1, he said, referring to a computing parameter that is only that inexpensive for AI systems with a small number of parameters. 

Adopting AI at scale is “even harder in the enterprise” world, Mr. Nilekani said, as companies are far more cautious about handing over decision-making to models that are prone to errors. Mr. Nilekani was speaking at Carnegie India’s Global Technology Summit.

“But the most difficult actually is implementing AI in the public sector,” Mr. Nilekani said, “because the public sector has structural concerns. It has ministries, it has departments, everybody is territorial… So data is not shared. If data is the lifeblood of AI, we have to find a way to bring all AI together, irrespective of which part of the government it comes from.”

Indian models

“DeepSeek has done the world a great service,” Mr. Nilekani said, referring to the Chinese firm that put out a reasoning AI model at a far lower cost than western competitors. The firm “accelerated the commoditization of models,” he said.

On building an India-specific large language model, he said that nobody “should lose sleep” over an indigenous model not being built. “The real challenge is how do we make this work? How do you make it work for a billion people? How do you make sure it’s accurate? How do you make sure there’s no bias?”

“So we’ll focus on those things tomorrow. If some new model comes, we just replace the old one with the new one. No big deal.”