How do you see India’s developer community furthering the cause of AI in the country, and helping Microsoft achieve its goals here?
The fact that India (with 13.2 million) is now second only to the US (about 20 million) in terms of the number of developers on GitHub and expected to cross the US in 2027, is (a) tremendous (feat). The other thing that’s super exciting for me on this trip is to see the broad adoption of GitHub Copilot (Microsoft-owned cloud-based AI developer tool) and how it’s being used not just to build tools, but also to make the tools democratic by getting lots of people to use it.
This is just a taste of what India is capable of. This didn’t happen with the (other tech shifts such as) PCs (personal computers), client servers, the web internet, or even mobile cloud where India was a big user of everything. Now, India is a developer of all this, and that is the big transformation. My dream has always been for India, and every emerging market, to be on par with any developed country when it comes to exploiting the benefits of new technology. Economic growth is not about just consuming tech, but also about creating tech—and that is the opportunity.
You have been at the helm of Microsoft for a decade and have overseen major technological shifts. What next?
As I said earlier, this is my 32nd year at Microsoft, and I’ve seen these four big platform shifts—the PC client server, web internet, mobile cloud and now AI. The big thing for me is as a company to be grounded in how we can empower people in organizations, and then to have a culture that’s learning. In tech, we’re only as good as what we do next, (and) not (rely on) what happened before.
What skills should developers build, in the backdrop of the omniscient AI?
On one side, there will be more big model breakthroughs at the intersection of electrical engineering and computer science, which will require some very hard science that India is well known for. On the other side, learning new frameworks, new languages and new abstractions is what makes one a great developer. Think of AI as just another abstraction that can be used to build your app.
Also, when you think about a development team, it’s a pretty inclusive one—you have designers, and now you can even have users participate in the development like the Shiksha Copilot example where you have a teacher working with a nonprofit (Sikshana Foundation), a software developer and Microsoft Research (India), effectively as a single team.
Ultimately, it’s not about just one engineer. Rather, it’s about engineers embedded in the real world with empathy for the world they live in that also has users in it.
With GenAI picking up momentum across sectors, how should CXOs view the shift of many applications from classical AI and GenAI?
There are two approaches one can take. The good news is that India is benefiting tremendously from the rails that were laid out during the cloud era, which has helped accelerate the AI era because you already have cloud infrastructure, which is elastic. This helps to get going on any project fast, since you have all your data, and even if you have legacy systems, many of them would have been migrated to the cloud. That sort of makes an AI POC (proof-of-concept) anytime-to-value, much faster.
From a CXO perspective, you can now have anything that is customer-facing (chatbot, voicebot, etc.) in Indic languages and with front ends (user interfaces) that are truly natural, which will allow even the vernacular rural population to be able to interact with your service or product or whatever you have. That’s a big breakthrough, and the benefits are immediate. Or they (CXOs) can even think about how can I have better customer service? Those are the top-down easy use cases.
The more horizontal one (use case) is to use Copilot to, say, increase the knowledge turns (learning cycles generated by experience that benefit others, too) in one’s enterprise. Axis Bank is an example. They were early to the cloud. Now they’re early to Copilot (first bank in India to adopt Copilot for Microsoft 365 at enterprise scale, with a little over 300 users). That’s what going to happen, where companies will say: we want new productivity, new ways to work, new ways for people to be empowered to change how we work.
That said, given the misuse of AI and GenAI for misinformation, deepfakes, voice cloning, hallucinations, copyright violations (eg. the copyright infringement case by The New York Times against Microsoft and OpenAI), etc., what should CXOs keep in mind before scaling up GenAI projects?
That’s why we love the design metaphor and came up with Copilot to insist on ‘humans in the loop’ (when dealing with) the first draft of a code or document or spreadsheet or whatever, (since) that’s where fundamental mistakes occur. You can now start with a draft and edit it and correct it. And hallucinations too, sometimes, are features and not bugs, which humans can figure out.
As for your point about copyright (violations), we believe every country and the world needs to decide how to use the ‘transformation’ part whenever there’s a new transformative technology—eg. deliberate on what is fair use, following which you can use copyright protection. And that’s something which will get litigated and get settled. Meanwhile, whether it’s a developer using GitHub Copilot or a company that is using Microsoft Copilot, we want to indemnify a customer anywhere, so that we do not stop the benefits of technology. The worst mistake we can make is to take some new general-purpose technology and shelve it because of what is something that I think is absolutely solvable.
But, again, CXOs may be getting unnerved by the misuse of AI such as deepfakes, voice cloning, etc., prompting them to wait till the technology matures before scaling it up in their companies. Would you agree?
I think there’s a healthy dialogue and debate that is happening so that we are able to deal with both the benefits of any new technology and their unintended consequences—whether it’s election interference, whether it’s deepfakes, or what have you. I welcome that and, therefore, it all boils down to what are the societal norms that can be translated into some law enforcement (for AI safety), similar to how tackled the botnets in the past with societal norms.
Microsoft has been focusing on the local language ecosystem in India. In this context, please share your thoughts on how India large language models are shaping up?
It’s fantastic, and I talked about what Sarvam is doing (developing LLMs specifically for India—the OpenHathi series). The next application that’s being written doesn’t have to wait for Indic languages and voices to be available. This, by the way, is so critical in India where people will want to speak in multiple languages and voices (and use) a low-latency (minimal delay) interaction model. The other thing is to use the ability of something like OpenAI, or Azure OpenAI (Service) with Sarvam as a developer and build a front end (interface) to sweeten what have you. The ability to conceive these applications that can take advantage of all of these innovations is what is going to be most impactful.