India uniquely positioned to drive new AI innovations: Google DeepMind

Two senior executives of Google DeepMind explain the laboratory's work in the country

Scientists and researchers at Google DeepMind, the Alphabet-owned research laboratory, are building new artificial intelligence (AI) systems. The technologies they are working on will advance science, transform work, and improve billions of people’s lives, said two senior researchers of the lab in an interview with Business Standard in Bengaluru.
 

“Google DeepMind’s mission is to build AI responsibly, to benefit humanity,” said Seshu Ajjarapu, senior director at Google DeepMind. 

Ajjarapu, who lives in the United States, was in India recently for the Google I/O Connect Bengaluru 2024 event. “What you see today are near-term (innovations). But we have our eyes on the longer term, where we are working on some of our time's most challenging and canonical problems,” he said.

Google DeepMind combines two leading AI labs — Google Brain and DeepMind — into a single, focused team. In the last decade, the teams have made the biggest research breakthroughs, many of which underpin the AI industry of today.

One innovation that Ajjarapu highlighted is AlphaFold, an AI system that accurately predicts 3D models of protein structures and has catalysed a new wave in biology. Google DeepMind recently unveiled the third major version of AlphaFold that will help scientists design drugs and target diseases accurately. 

Researchers have long sought a source of clean, limitless energy. Nuclear fusion, which powers the stars of the universe, is one solution.

Smashing and fusing hydrogen, a common element of seawater, releases huge amounts of energy. On Earth, scientists have recreated the extreme conditions of stars by using a tokamak, a doughnut-shaped machine surrounded by magnetic coils and used to contain a plasma of hydrogen that is hotter than the core of the Sun. However, the plasma in a tokamak is unstable and makes sustaining the process required for nuclear fusion a complex challenge.

To solve the problem DeepMind collaborated with the Swiss Plasma Center at EPFL to develop the world’s first deep reinforcement learning (RL) system that autonomously predicts how to control the coils and contain the plasma. It has opened new avenues in nuclear fusion research.

“India with its mobile-first population and booming startup and developer ecosystem…is uniquely positioned to drive the next generation of AI innovation,” said Ajjarapu, citing innovations made by Indian companies. From consumer experience to agriculture and social enterprises, AI is addressing the biggest challenges in sectors and industries.

Manish Gupta, director at Google DeepMind, said inclusive AI is a key theme of the company’s team in India. “We develop AI in a manner that brings benefits to billions of people, especially the ones not being touched by AI,” he said. “A lot of our work is on language inclusivity and multicultural awareness to make all of these models understand different cultural nuances.” 

The Google DeepMind India team spoke about their work in language solutions that help computing developers working in AI. This includes expansding Project Vaani – done in collaboration with the Indian Institute of Science (IISc) – which provides developers with over 14,000 hours of speech data in 58 languages. Gupta said the project is in the middle of Phase 2, expanding to cover 160 districts in Indian states. 

DeepMind, as part of its commitment to responsible AI, tracks biases that computing models may carry. It has made some data sets and benchmarks public so that other researchers can study the biases. “And then of course, we come up with mitigation techniques for our language models to get richer cultural understanding and not perpetuate these biases,” said Gupta.