Google DeepMind's Manish Gupta on building AI that can benefit everyone

Recognising that for many Indians their smartphone is their only digital device, Gupta challenged his team to make AI models work within the constraints of mobile hardware.

“This led to a lot of work in making these models efficient, and work with constraints like a certain amount of compute and memory. Now, we have Gemini Nano that is meant for smartphones,” he says. The same principles of efficiency have become critical in the large language model (LLM) era, where the cost of computation has ballooned.

View the full list here 

“Gemini Flash is currently the most efficient model in the market, in terms of rupees per watt. I’m proud of our team’s work in contributing to the same.”

Beyond foundational models, Gupta’s team has applied AI to public health and agriculture. In collaboration with the non-profit Armman, they developed models to identify expecting mothers at risk of dropping out of health programmes—improving engagement and outcomes. In agriculture, they built the first AI model to map field boundaries and crops across India using satellite imagery, enabling farm-level insights for loans, insurance and subsidies.

“Manish is a rare, passionate leader who bats for his team, and gives them enormous creative freedom. At the same time, he does not hold back if the work does not have potential impact,” says Shiv Kalyanaraman, CEO of ANRF. “Manish and team have always put impact and helpfulness for a billion people at the heart of their work.” 

Also read: AI futurist Harrick Vin foresees a world in which humans and AI help each other

Looking Ahead

Gupta’s team has also been at the forefront of addressing bias in AI, particularly in non-Western contexts. “We published one of the first papers identifying bias based on religion and caste in Indian datasets,” he says. “We’ve been publishing papers on how many of these biases also carry over to visual models when you’re trying to understand images. Most of the AI models lack a cultural understanding.”

To address this, the team has been working closely with researchers across India, including institutions like IIT-Madras, to improve Google’s models and support the broader ecosystem of responsible AI development.

Gupta is excited about AI’s role in agriculture, health care, accelerating science and drug discovery, among many others. He cites AlphaFold, which won the 2024 Nobel Prize in chemistry, for its contributions to protein structure prediction and computational protein design. “It can predict the structure of 200 million proteins in under a year—work that once took a PhD student five years per protein. That’s a billion PhD years of work done in less than a year,” he marvels.

What does he hope his legacy will be? “Inspiring and challenging teams to go after ambitious goals—not settling for a modest impact, but solving real-world problems that make a difference to humans.”.