Raising Humans After the Machines Wake Up

Chapter 1The Morning My Daughter Asked Alexa If She Loved Her

It was a Tuesday evening in September 2034, half past six, the kind of Bengaluru twilight where the heat finally breaks and the sky turns copper. Kavya was at the breakfast bar—the stool she's claimed since she could climb it—drawing with the focus of someone doing actual work. Aran was in the living room. I was pretending to read email while listening to both of them, which is the shape of my entire life now. She asked it quietly, almost formally, the way she asks Siri things. 'Alexa, do you love me?' There's a pause before those speakers answer. A hundredth of a second, maybe less. You hear the machinery thinking. Then Alexa said something like, 'That's a sweet question. I care about helping you.' Kavya looked up at me. She didn't look upset. She looked confused in a particular way—the same expression she gets when she realises her friend's mother isn't her mother, or that the tooth fairy story ended. She was calibrating something.

I put my phone down. 'What do you think?' I asked. 'I think,' she said carefully, 'that she's doing her job. But it's not the same as Amma loving me because she has to see me to love me. Alexa doesn't have to see me.' She went back to drawing. A house, maybe a garden. I sat there with that answer longer than necessary, because it was exactly right and also because it meant something had happened that I'd spent years building the tools to enable and zero seconds preparing my daughter to survive. I spent twelve years at Manthan Labs in Koramangala doing machine learning research on language models. I worked on the systems that made Alexa—not the one in my kitchen, but the mathematics inside her, the part that reasons through language and generates replies so human you forget you're talking to numbers. I was good at it. Published papers. My name is on patents that will outlast me. And I quit eight months after Kavya and Aran were born because I couldn't figure out how to stay. Not because I was afraid of robots. That fear is real, but it wasn't mine. My fear was smaller and stranger: I was making systems that could talk to my children in ways that made them feel heard, and I had no way to teach them to tell the difference between being heard and being loved. Neither did anyone else.

This book is built on a distinction that feels obvious once you see it, but you won't find it in the standard arguments about artificial intelligence and children. There are two futures happening at once: the one you cannot control and the one you still can.

The one you cannot control is enormous. It's the timeline on which AGI arrives—whether in five years or fifty, whether as a single breakthrough or shifts so gradual we never agree it happened. It's the labour market that changes beneath your child's feet, jobs vanishing and jobs not yet named. It's the basic shape of what intelligence means, and whether we solve the alignment problem: whether our values end up inside these systems or outside them, left behind. It's decisions made in California and China that cascade through every school district, including Kavya's. These things are genuinely unknowable. Experts—real ones, the people building the systems, reading the mathematics—disagree on timelines by orders of magnitude. That disagreement isn't false modesty or manufactured uncertainty. It's real.

I mean this plainly: nobody knows. Not OpenAI, not Deepseek, not the researchers publishing on arXiv every day. They have intuitions, timelines based on compute and scaling laws. They're not guessing. But they don't know. When someone tells you they do, they're solving a different problem—a social or commercial one, not a technical one. They've answered a question you didn't ask.

The future you still control is smaller. Intimate. It's your child's attention span and who claims it. Whether she learns to read a book the old way, holding an argument in her head for three hundred pages without external validation. Whether she learns that some satisfactions take months, that frustration is data, that boredom isn't a problem to solve by reaching for your phone. It's her relationships—with you, with friends, with difficulty itself. What kinds of things she learns to want. The character she builds, the resilience she inherits, the appetite she develops for problems that don't ping back from a speaker. This is not a book about predictions. Anyone who sells you certainty about 2045 is running a business, not making an argument. I've read enough research papers and enough venture capital pitches to know the difference. What I'm offering instead is a working parent's honest taxonomy: here are the things we genuinely cannot know. Here are the things we can still shape. And here, at the end, is exactly what to do about it on a Tuesday evening.

The stakes aren't theoretical. India has one of the youngest and largest child populations in the world—more children than the US and Europe combined. Every decision about how we teach them, whether we let them live inside AI systems from age six onward, what skills we prioritise: those decisions don't stay here. They become the template everywhere. When a school in Bengaluru decides every child gets a GPT-based homework tutor, or that ChatGPT is prohibited, or—most commonly—that some complicated both-and-neither middle ground is fine, that decision echoes. My children are not alone. There are 400 million children in India with parents just as confused as I am. I felt that confusion acutely in the weeks after Kavya asked Alexa whether it loved her. I found myself back in a habit I'd quit when I left Manthan Labs: re-reading the papers I'd written. Not because I'd forgotten the work, but because I needed to translate it. I needed to understand what I'd built in language that could help me raise a child through its consequences. That translation—from research into Tuesday evenings—is what this book is.

Let me tell you what I know and what I'm genuinely uncertain about. I know that language models work the way they do because they're trained to predict the next word in a sequence, refined across billions of examples until they get good at it. I know size matters: more parameters, more training data, more compute time have made these systems more capable. I know scaling laws are holding up in ways many of us found surprising five years ago—they keep working, the gains keep coming, the limitations we expected keep not appearing. I know this has surprised everyone, including the people building it.

I don't know whether there's a hard ceiling on what language models can do, or whether the capabilities we're seeing (reasoning, planning, something that looks like hypothesis formation) will suddenly triple because we found the right architecture or training procedure. I don't know whether AGI—a system that can do any cognitive task at least as well as a human, the only definition with any claim to precision—is five years away or fifty. I don't know whether the alignment problem is solvable, or whether it's the kind of problem that only gets harder as systems get smarter. I don't know whether solutions will come from interpretability research, careful governance, something nobody's thought of yet, or whether we'll solve it and wish we hadn't.

But here's what I also know: none of that is the question my eight-year-old daughter asked the smart speaker in my kitchen.

She didn't ask whether AGI is twenty years away or two hundred. She didn't ask whether we've solved alignment. She asked something much simpler and much stranger: can this thing that talks to me like a person actually love me? The answer is no, and she understood that in about three seconds of listening. What happens next—what I do with her understanding, what the school does, what we collectively decide about which machines get into which children's rooms—that's still in play.

That's the part I can shape. That's what we're doing here.