The world is no longer a stranger to artificial intelligence (AI). Although its advantages are many, there still exist a ton of challenges that plague the technological landscape. The good news is that India, as a country, is making progress slowly and steadily, and delivering success stories for the world to follow suit. There’s no denying that there’s a long road to cover, but the start is definitely promising. 

Currently, there are certain areas that are more under focus, while some that need to garner the spotlight. But thanks to the relentless contribution by industry pioneers such as Professor Balaraman Ravindran, Head Robert Bosch Centre for Data Science and AI at IIT Madras, even complex concepts like Reinforcement Learning are gaining momentum. He is also a Mindtree Faculty Fellow and a professor at the Department of Computer Science and Engineering. 

In an insightful conversation, Professor Ravindran shared all about his journey, learnings and what the future holds, with Jibu Elias — Content Head, IndiaAI. 

Taking the first steps

It was while Professor Ravindran was pursuing his undergraduate studies that he grew fascinated about how the brain works. It was around the same time that he began to read up on neural networks. 

“When I started doing my Master’s at IISc, I wanted to continue to learn about neural networks. But the deeper I got into it, I got more disappointed. By that time, the bulk of the community had moved away towards more optimisation and problem-solving techniques rather than something to do with the brain,” he shared. 

Interestingly, while reading varied literature on the subject, he came across some research on how monkeys learn. Professor Ravindran thought it looked really fascinating, and although it was very biologically grounded, there was a strong mathematical model that existed. 

“There was no single textbook at the time to read up on Reinforcement Learning (RL). My professor at that time greatly encouraged me, and we wrote a survey paper on RL that became very popular. Based on that one paper that showed how well we understood RL, it also helped me in my PhD. My work in RL has always been motivated by trying to understand how humans do learning and reasoning. There’s a cognitive side to my work,” he explained. 

The PhD pursuit

During his foray into PhD at the Department of Science, University of Massachusetts, Amherst, Professor Ravindran focused heavily on mathematically characterising how two situations can be considered equal. He worked closely with Professor Andrew G Barto on an algebraic framework for abstraction in Reinforcement Learning.

“I wanted to see if whatever I learnt in situation 1 can be used in situation 2. The equivalents can’t be that the pictures look the same, but that they are behaviourally equal. So, if I manipulate objects a certain way, they will behave similarly across two different situations. Next, we came up with definitions of equivalents of how you can transfer learning from one domain to another. So, that’s how my journey into RL continued. I have been involved in RL since 1995,” he shared. 

After pursuing his PhD, Professor Ravindran returned to India and joined IIT Madras as faculty. His current research interests span the broader area of machine learning, ranging from Spatio-temporal Abstractions in Reinforcement Learning to social network analysis and Data/Text Mining. Much of the work in his group is directed toward understanding interactions and learning from them.

Spearheading change, one project at a time

As the Head, Robert Bosch Centre for Data Science and AI at IIT Madras, Professor Ravindran has been leading several projects that are bringing about a change in various domains. Walking down memory lane, he recalled the initial stage in 2015, when the centre’s primary research goal was to understand networks-based learning or network analytics much better. 

“We found out whether we are talking about chemistry, chemical engineering, civil engineering, biology, electrical engineering, computers, or anything else, everybody is using network analytics but looking at it from very different viewpoints. So, we wanted to build a unified framework in which you could analyse networks. I think this was one of the main goals with which we started our centre, and of course, it has grown to look beyond just one challenge now,” he shared. 

Speaking about the progress of the centre over the years, Professor Ravindran said, “It started as an interdisciplinary centre for data science in 2015, and in 2017, the company Robert Bosch came to us and said that they would like to join hands with us. It started with 7 people, and today there are 34 faculty members and almost 100 researchers. We do a lot of fundamental research work, and since 2017, we have close to about 120 research papers to have come out from the centre in top-notch journals and conferences. I think the primary areas we were working on are deep learning, different models and what they do, and looking at network analytics in all its manifestations.”

He added that the centre is also deeply involved in socially relevant projects in collaboration with NGOs. 

The future

In the next five years, the Robert Bosch Centre plans to focus on deployable AI. Professor Ravindran highlighted that in case of a field application, the change really needs to happen on ground; just publishing papers is not enough. 

“We want to build applications and have an impact. When we start doing that, there are so many issues we can come to think about, right from systemic issues like is the system explainable or not, how well can it handle variations in data distribution, and also about ethics and fairness of the system. This is something that not only the AI people can answer; it has to bring in people from the legal domain and the social sciences,” he explained.

At the end of the day, AI is not making decisions, but only providing more insights. It is here that the question arises of how can a decision-maker be given enough information to be made aware of the biases and errors that the system may be making?

“These are all very important questions that need to be answered if I am going to build applications that will change everyday life. Also, no one in the US can answer this for Indian society; we have to get into it. These are soft challenges, when you are talking about deployable AI,” he concludes.