Tech regulation should be done in national interest: Satya Nadella

The company’s $1 billion investment in Open AI LP, the for-profit arm of Open AI Inc.—the buzzy non-profit behind Chat GPT that is all the rage in the tech world right now, is being seen as giving Microsoft an advantage in fusing AI into search to usher in new ways in which we seek information online.

During his ongoing four-day visit to India, the Hyderabad-born Nadella sat down with Mint for a wide-ranging conversation on AI, tech regulation, startups, modern work, India’s public digital infrastructure and how he views his run at Microsoft. Edited excerpts:

Are you looking at Bing launching with ChatGPT or another language model built in?

We are very excited about the work we’re doing with OpenAI, and you can expect us to do a lot with foundational models. Like what we did with GitHub co-pilot was an early taste of what I think can be the fundamental change in what happens when large foundational models and products like GitHub come together to empower people. A software developer today is able to be so much more productive, creative, and able to really flatten the learning curve. We want to do more with AI, quite frankly, across the length and breadth of all Microsoft products. So, I would say wait and see because we are just getting started.

In terms of accuracy in search results, are these language models really there yet?

I think search and large language models will be brought together in very novel ways. There are fundamental challenges around any model—about how you ground the model and how you keep the freshness of the model—those are all good technical challenges that I think will have good technical answers. And in due time, you will see the output of all of that.

Because the English language is what such models are mostly getting trained in, what happens in countries like India, where there is a multitude of languages?

That’s a great question. I think we should not think of it as just this one shot. It’s the continuous refinement of the model, with human feedback and content that the AI generates. So, suppose I wanted to read Lorca translated into Hindi or English. I can do that today. Is that then a Hindi poem? Is it English transliteration? Is it a Spanish poem?

What is considered knowledge in one language can be consumed in another language and perhaps modified in another language, which then makes it back. So we should think about the fact that more all the people in the world consume these large models in all languages and add to it; that’s how we truly get to democratize these differences and also biases. Ultimately, collective human judgement is the only test against bias.

What do you think people do not fully understand about where AI stands today?

I would say the thing that perhaps needs to be better understood is some of the scaling effects and the emergent behaviour of large models. I’m not saying that this is the be-all and end-all or the last development there is going to be of model architectures.

But GPT3 or 3.5 are not linear steps. So the question is, where is the nonlinear advancement coming from? There are these studies where if you train a large model on a corpus of, let’s say, all math equations. It gets good at that. But if you train it on all math equations plus all web text and all of the world’s literature, it gets better at math than just the corpus that it was trained on.

And you could say, why is that? That’s less intuitive because you and I didn’t go to school and only learn math. We learnt history, we learnt languages, and we learnt a lot of other things, which, by the way, created the general circuits (in our brain) that helped us get better at math. So that, I think, is the key.

In the evolution of India’s tech industry, we have large legacy tech services firms. We have more recently seen a wave of consumer startups. Are we about to see another wave of startups that exploit AI and the cloud?

One of the things that give me great optimism for India, quite frankly, is all the different ways in which technology is getting used. I got to spend time with a small business called Senco Gold which is expanding its retail stores using tech. I met a gentleman at the State Bank of India who is visually impaired and was building an app using Power Apps to automate a set of workflows.

India already has the second largest number of developers, and is No.1 on AI repositories, which means Indian developers are seeking out every open-source AI repository and becoming core committers. So, the simple answer to your question would be, in the long run, there will be a lot more intense use of AI. One of the things that you should talk about, even in India, is the intense use of new technology, whether it’s a gentleman at SBI using AI or the latest unicorn, because both are super valuable for India’s development.

As the super boss of both Microsoft Teams, as well as LinkedIn, what can you tell us about how work has changed in the post-pandemic world?

There are three data points in terms of the post-pandemic world that we are all learning. One is there is this paranoia about productivity. Some leaders say we are being productive enough, while employees say we’re getting burnt out because am I working at home or sleeping at work? I think the way to bridge that is through data. Have aligned outcomes. Every organization has to be competitive, and, ultimately, work needs to get done. Using that versus being dogmatic about any particular way the workflow worked pre-pandemic is probably useful.

The second data point is people come for other people; they don’t come for policy. I think we all as leaders have to learn some soft skills on how to convene people together. Learning better soft skills on how to create connections, I think, is important.

The last thing is, just because you recruit them once doesn’t mean you don’t get to re-recruit them every day. None of us can take anything for granted, especially in a market like India.

What role do you see India playing in the overall regulation of technology, and have our current regulations affected your business?

I think every country is going to have regulations so that the technology that’s both been brought into the country and that’s been created in the country ultimately serves its social purpose. We welcome it and also are not waiting for regulation in order to have a fundamental trust in technology and business models of technology. That’s how I approach it.

It’s a continuous evolution of regulatory regimes and technology, and things have to keep pace. And just because the regulation cannot keep pace with the technology that’s going to come, it’s incumbent, quite frankly, on people like us who are in this business to say how we have core foundational trust in what we build, by design.

Some of what I’m seeing in India, even on data privacy and data flows, is very enlightening. Ultimately, India has three things. One is we are making in India by building our data centres and human capital so that others can make more in India—whether it’s Senco Gold, SBI or InMobi.

But they’re not just making in India for India; they’re making in India for the world. So then you have to have regulation that accounts for all that. Data doesn’t come from one place. It may have gotten created somewhere else, and it should have come to India. Then in India, people should be able to use it, add value to it, and then have it go out of India and have others use it. That’s the way for regulators to think about it. And by the way, it should all be done in the national interest. It’s not about benefiting anybody else other than the people of this country.

When you took over as CEO, you had very large shoes to fill. Now you’re being spoken about as among the great CEOs of the world. How do you look back at your tenure at Microsoft?

I feel that if I have to live the culture that I espouse, I have to be more focused on the mistakes I make each day so that I can learn from them and get better tomorrow. And if anything, all of us are temporary stewards of institutions that hopefully, if we do things right, should far outlast us.

If there’s one thing that I learned from my Indian civil servant father, it’s that the institution being stronger after you leave more important than whatever you achieved during your tenure.

I want to make sure that I, personally, and everyone at Microsoft has the confidence to acknowledge our imperfections.