At the NVIDIA AI Summit in Mumbai last week, Jensen Huang, the chief executive of the one of the world’s most valuable companies, made it clear that it falls to information technology giants like Tata Consultancy Services (TCS), Infosys and Wipro to usher in India’s artificial intelligence (AI) revolution. Amid the din of legacy players and start-ups talking up their AI plans at the Jio World Convention Centre, TCS’ AI supremo Sivaraman Ganesan sat down with us to talk about the Tata crown jewel’s blueprint to survive and thrive in the age of AI.
The Tata Group has led India through a number of technological shifts—from steel to hydel power, to semiconductors. Would Tata also build India’s foundational models in AI?
Advertisement
We are watching this space closely. If you ask me if we are looking at building a foundational model at this point in time, the answer is: no, we are not.
In the future? Maybe. If a need arises. Currently, our play is to look at the open source and other models that are out there and extract business value for our customers. That is the space we are in.
It was apparent from the NVIDIA chief executive’s speech that he sees IT services majors leading the India AI story more than tech start-ups. Does that surprise you?
I think Jensen Huang used the word tectonic shift—the all encompassing change that is happening in the world with AI. The companies who are being born now will grow up as AI native or AI first.
We read headlines every day that start-ups will have a role to play. But then, we must realise that society also comprises companies that are tenured. They all have a variety of older generation technology in their computing—from the 80s, through the 90s and the mid 2010s to the early 2020s. All of these need to be now consuming AI.
That is where our role comes in. While we infuse AI in the older stuff, we also help them in building new things ground up on AI.
Advertisement
Yet, many analysts and tech leaders are sounding the alarm over the possibility of AI disrupting the IT services sector. What is your evaluation?
We have been through several evolutions of automation over the past 30–40 years. And through this time we have kept finding that some or the other part of work can be automated. As Jensen Huang said, the competition is going to be between a person and another person who uses AI. I think this [AI] is an efficiency game with exponential outcomes in store. It can only mean that our skill-level is elevated by the consumption of AI.
Do you think AI can help in intellectual property creation and product-isation?
Historically, we have had our share of products. You would have heard of products like BaNCS, Optumera and Digitate. They will benefit from AI.
Will AI also allow us to ideate and create new intellectual property? Absolutely. Do we see it as value additive—with the combination of data, AI and the cloud coming together to create even more value? That is a huge yes. That is the reason we created the AI[dot]cloud business unit.
It seems the first and most significant implementation of generative AI is happening in customer support automation. Is this true, and what kind of savings will a company make if it implements AI into customer support?
Advertisement
Our vision for the future is that a lot of customer support will be proactive rather than customers having to call in, complain or raise a ticket. The advantage of AI is that it can anticipate problems and reach out to customers proactively in an outbound manner. The ultimate vision is not only to tell the problem, but fix it.
[On AI’s capability for customer support], I would say it is upwards of 8 or 9 on a 10-point scale. In terms of savings, you should be seeing more than 20–30% for starters. But it is difficult to quantify this because the game is just beginning. As the market and technology mature, we will see improvements. So, it will not be good to hold these numbers.
When it comes to generative AI, we have to put in extra guardrails and do much more of risk evaluation because any mistake can dent your reputation and brand. We have to be careful.
Advertisement
One of the biggest factors of production in AI is data. Experts say the most significant challenge is to gather and streamline data from silos before implementing AI. Are you seeing a demand for such services?
How you assimilate large volumes of data, how you aggregate it, how you use it for combinations of models and then extract intelligence is a very big business. And we are seeing a lot of demand in the space.
Many people talk about AI but not as loudly about data. And data is supremely important for AI. One thing is to have the models trained and ready. But more importantly, once you consume a model, how do you use inference techniques to augment the model to fine tune it and get the actual business information you want? That requires quality data. By quality data, we mean data which is accurate and representative of the enterprise.
Also, the right privacy and security guardrails must exist for the data. There also has to be explainability—once you get a business result, AI must be able to say how you can verify where it came from and how it was ingested and transformed.
Advertisement
It is said that generative AI is a black box—we don’t exactly know why a particular output is produced for a given input. Have you found it possible to actually trace the explainability?
It is an evolving space. I will not be able to say yes or no right now. More and more of our regulatory rules are starting to get defined and technology will have to conform to the laws of the land.
You are right about the black box phenomenon. But I think as people who implement technology, that is where the guardrails come in. Otherwise auditors and people who are tracking risk will not be able to prevent something from happening.