Rohin: But as it turns out, maybe there’s a silver lining. I was just chatting with Chaitanya yesterday… and we were kind of joking about it wasn’t a bad thing that India was late because, what did we really miss if you look at it in some senses… Let’s say we were early and you spent a bunch of mostly taxpayer money… what would we have gotten? Because now the AI world seems to be arriving at a new reality.

Chaitanya: The new reality is that there is no AGI… The original hope was that… there will be a system which is the AGI, right? Artificial General Intelligence where it has learned everything, and any new things also it’ll learn on the fly quickly, and it will be able to do any job that a human does better… I don’t think that at least the US has given up on that yet… But at least even in the US, most of the researchers… agree to the fact that AGI is no more a dream—at least with the current technologies.

PGK: Why is it now the conventional wisdom that, you know, AGI is probably not going to happen?

Chaitanya: It is conventional wisdom now if you look at whether even Yann LeCun and Gary Marcus and everybody… in research circles accepts now that the current LLM models and the architecture are not enough… because they have not been able to learn the world models, as we call them. And AGI is not imminent, and in fact, it will not happen without other breakthroughs.

Chaitanya: GPUs—if I have to, first of all, invest and buy anything, I’ll have to get it from Nvidia itself. And it’s costly. And it’s a one-time thing where if I pay for an hour, that hour is gone and I won’t get it back… Everybody knows that to build models, the most important thing is the data set. Other than the data set, the architecture is known, the training parameters are known, the hardware is known—everything is known. You just need to plug in the data sets, and we don’t have the data sets.

Rohin: What is India’s AI strategy? …my proposed 2X2 matrix… On the X-axis, I have on the left, ‘Capital and scale.’ And on the right side of the X-axis, I have ‘Ecosystems and ssage’… My Y-axis has, on the bottom, ‘Government and regulations.’ And on the top, I have ‘VCs, PEs, and sovereign funds.’

Chaitanya: The people who got the money from the India AI mission are saying that they have to train on synthetic data because there is no data. And that’s such a useless way of doing it… Synthetic data is where you ask chat GPT to generate data in Telugu. So where is the culture? It’s all gone. It’s like a cyclical effort where you’ll end up reinforcing the same biases.

Rohin: What can we do today that will make the world look very different three years from now? Very few people, right? But doesn’t that same argument hold? If India solves that problem today, next year, or the year after, imagine how different the world will look three years from now or five years from now.

Chaitanya: We simply ask them, ‘Go and ask ChatGPT to tell you a story in Telugu.’ Or I say, ‘Explain fractions in Telugu.’ See, the examples it picks up—let’s say it gives an example of a pizza being cut into three pieces out of four pieces or whatever… Let me tell you, if you go to the hinterlands, the rural areas, nobody knows what a pizza is. There is no way the context… translation is not just translation, you’ve got to translate culture.

Rohin: Imagine a situation where American kids or Americans were reading examples about poha and smoking beedi and wearing lungi in English. It would make no sense to them. I think we’ve been so conditioned over time to the… American cultural context that, especially us in urban India, we don’t realize how out of place it is.

Chaitanya: India is an oral culture. Most of us, we are not written… We don’t have a bedtime story-reading culture. We don’t say that, ‘Hey, read this story.’ We say, ‘Dadi, mujhe story batao.’ So, she tells the story… When we say language-poor, low-resource languages, which we call—that’s actually a misnomer.

Rohin: India is being culturally excluded from the AI world that we’re moving into, where data is everywhere, and we are sort of being culturally excluded. And it’s not that somebody’s standing at the door and saying you can’t get in; it’s just that we aren’t getting our act together.

Chaitanya: Learning for a colonial child became a cerebral activity, not an emotionally felt experience… The teacher herself is converting the concept in her mind from her mother tongue into English. And then I listen to it in English, and then I try to convert it into my mother tongue. And then the whole thing is lost.

PGK: I’ve never heard this argument, that you know what, because we have made our humans into drones and machines, we are best placed to create the agents and machines that are going to replace them.

Chaitanya: Prove me wrong. Who can do it better than me? Who can do it better than Wipro, who is sitting on—every day they generate 10,000 employees’ data, like what to click, what to tell, what log to check?

Rohin: Running data centres or outsourcing centres for decades has generated a wealth of process insights. And it’s those process insights which can essentially power the next level of automation and, quote-unquote, ‘agentic’ innovation.

Chaitanya: Now the point is that now that resource, the original agent, can be… that’s the original agent, yeah. So now we are best placed to actually build those agents. If you really think about it, we have the prompts, we have the training material which is the prompts, we have the data to train… The pitch they have made is that AI will replace humans. The pitch we are making is it’s going to augment humans.

PGK: I’m assuming it’s like the ultimate form of innovator’s dilemma, because now the thing is that if you have to spend your time and effort to do it, it means that you have to reconcile yourself to the idea that… you will have to fire or get rid of a large part of the people who are there.

Rohin: What people don’t realise is that it’s going to happen anyway. You can either decide to do it to yourself, or someone else can do it to you.