‘Auditing AI is the key to the future, the machine should be able to explain its decisions’: Madhusudhan KM of Mindtree

IT Services companies have new opportunities in the era of AI, they are well poised to deal with complexity in merging the digital world with the legacy world. But, they have to be fast, focus on talent development and start collaborating with startups.

Madhusudan KM, CTO of Mindtree, does not have an easy mandate. He is tasked with steering his company through some of the biggest technology challenges and opportunities of this time. Artificial intelligence (AI) and machine learning (ML) are here to stay and they are going to impact all our lives. In that context, he talks of how the world has to grapple with going digital and why IT Service companies play a special role in transforming companies in the era of AI. He notes that India’s biggest challenge would be in getting Indian talent up to speed in these modern technologies, and emphasises that industry-based learning is the only way to bring about change in engineering education in the country. Here are some excerpts from his interaction with YourStory.

YourStory: What does scaling up mean in this digital age, a world of startups and corporate innovation?

Madhusudhan KM: Nine out of 10 startups that I have met are studying to scale. So it has been our objective: to see how we can partner with them and then jointly reach our customers. Knowing the strengths of their solutions has been the premise for all of our startup ecosystem conversations. Every company is struggling with digital transformation and how to define it. It’s different for each organisation. Thirty years ago, some people even called the mainframe digital.

So what is digital then? The key definition is how I can transform my business, which I have been doing in a certain way, using any technology. It doesn't matter whether we use mobile or cloud technology. Technology is just an enabler at the end of the day.

The problem with companies in the business for 30-40 years wanting to go digital is their legacy technology. These global enterprises are not good at replacing legacy, so they keep adding new technologies on top of the old. As a result, the whole architecture has become very complex in most of the enterprises. The organisations that will succeed in the future will be those that cut through the whole complexity.

So let’s take the airline industry, for example. Starting from the processes of booking a ticket, going to the airport and checking in, all the airlines today still use mainframes. If the core processes are there in mainframe, how do you then change the way you go digital? For example, even though I order through an app or through an aggregator, the actual booking is happening on the mainframe. So what we call PNR is actually stored in the mainframe. This is just one example.

All the airlines have tried moving away from mainframe to integrate modern technologies, which have gone on top of them. Huge investments have gone in but none of them have succeeded so far. That’s what I tell customers: we can use any technology as long as we can stitch it all together to serve the customer better. We, as an IT Services company, can deal with that complexity and start building more modern systems. For example, for big data systems, let’s say you want to understand customers and offer personalisation. We can still pull data out from the mainframe, use an application like SAP, and create a more modern data-lake on a modern infrastructure

YS: Yet, why are IT Services revenues from digital not very large?

MKM: because it takes time for the IT Services industry. For a company like ours, 40-50 percent of the revenues are coming from digital. But if you take some of our larger competitors, it is about 15-20 percent. Mid-sized companies like us definitely have advantage in changing directions. The numbers are one thing, but the technology is moving from being an enabler to becoming outcome-driven, and IT Services will build for that world.

If you take Tesla as a car, there are few mechanical components. Tesla is driven by software and lots of sensors. It is probably 70 percent hardware and software, with only 30 percent being mechanical. What we call a car has become a new platform for developers, because you can develop so many applications on the Tesla platform. The autonomous car is also not far away. That is the business, that’s the digital world.

YS: Give us some insights into what kind of digital transformation you have worked on.

MKM: We have done very interesting work with one of the retailers. Over the last 10 years, innovation in retail commerce has stagnated. We help them understand the future of customer shopping behaviour. A retailer launched a digital service with search playing an important role. But where was the intelligence? Today, retailers want to understand millennials. Let me give you an insight: I might follow a rockstar on Pinterest to see what he is wearing. If I am interested in the product, I will just take the photo, drag it and drop it on the website of a retailer and I should at least get a similar looking product. This can happen with multiple products. You see how buyer behaviour changes?

We are working with a B2B company that sells industrial boots and industrial parts to technicians and people into DIY. The case study here is, can the digital platforms guide these shoppers into not only what they are looking for, but also enable them to understand what else they could buy next? There are technologies like chatbots that understand key words to help customers and businesses. We have built a bot called Messi, which is an internal bot from which employees can find out how many leaves they have etc.

The technology we are building understands intent: you study 100 affluences, all of these need to have the same intent. This is also the foundation of AI. Today, we are teaching a machine to understand intent. It is like we are babysitting it and teaching it.

 YS: You have been talking about auditability of AI. Why?

MKM: We absolutely need auditability and explainability. There are two aspects: one is for serious enterprise-level AI adoption, for which technologists must ensure AI can explain why it made a particular decision. Explainability is very crucial. For example, we are working with an airline on a pricing determination algorithm. This is based on probabilistic and statistical models for now but the airline wants to go with ML and deep learning. As you know, every seat on a flight is sold at different price and, if unsold, it is a loss for the company. Every airline needs more than 75-percent occupancy for that flight to be profitable. Now, when the machine is deciding the price and the flight still runs at a loss, then the business leader should be able to ask the machine to explain why it determined that price which led to a loss. This does not exist today. The machine can make decisions today but you can't go and question it, and that’s what we want to make sure businesses have complete control over.

The other aspect I want to touch upon is how you can enable machines to make unbiased decisions. For example, it cannot throw up answers based on gender or race of a person. It is basically called as fairness in the eye. How do you train the machine to be fair in its answers? This itself is a heavy topic, the reason being human beings themselves are not fair today. So expecting a machine to be fair is a big ask.

YS: What are the challenges that India should gear up for?

MKM: Only challenge that I see for India is transformation of talent. I think our engineering colleges need to do a lot to opt in to the game of AI and ML. The industry will have a very important role to play and some colleges are actually very smart in creating partnerships. They have set up Internet-of-Things (IOT) and AI labs for their students in the last year and they put students through several projects. Many send their students for internships in the industry, but most of the colleges are still lacking. I think they are teaching the same things to the students that I learned 30 years back. Blockchain and quantum computing is something that needs to be taught too. So this is why we, at MindTree, have a centre of excellence to create a horizon where we ready our employees to help businesses transform. The second horizon is AI and cyber security.