Two young twin girls, around 6, sitting together on a sofa with a tablet. One points excitedly at the screen; the other sits with her chin in her hand, thoughtful.

"Papa, Why Should We Use Google and Not Gemini?"

My six-year-old twins had a debate about this. One wanted the answer. The other wanted options. The second one understood something most adults have not figured out yet.

The debate verbatim
Twin 1
"Papa, why should we use Google and not Gemini? Google gives so many links to read. Gemini gives me the answer."
Twin 2
"I like Google. It gives me options."

They are six. Neither of them has studied information theory, epistemology, or search engine design. One of them just articulated... without knowing it... the most important tension in how we interact with AI today.

One daughter wants the answer. One wants options. I sat with this for a while; because predicting what the world looks like for them in ten or fifteen years is genuinely hard. But one thing I am confident about is that logic will always matter. And for logic to work, you need something underneath it: a basic understanding of the data or information you are working with.

This is where the "Gemini gives me the answer" approach quietly breaks down. Let me give you a specific example.

The question:   When was Frozen 1 released?
Gemini's answer
November 27, 2013
Clean. Confident. One date. Technically correct.
What actually happened
Nov 19, 2013 World Premiere... El Capitan Theatre, Hollywood
Nov 27, 2013 Wide release in the United States
Nov 29, 2013 Released in cinemas across India
Why this matters: A child who only gets the AI answer learns that there is one release date. A child who searches Google learns that movies have premieres, wide releases, and staggered international rollouts... and that the same "fact" can mean different things depending on where you are. That is not a trivia lesson. That is how the world actually works.

The AI answer was not wrong. It was incomplete. And incompleteness delivered confidently, to someone who does not know enough to know it is incomplete, is its own kind of problem.

I saw the corporate version of this in Pune last month. A junior analyst spent three days and Rs 12,500 of his own money on a premium AI tool to generate a market map, only to proudly present a strategy targeting a competitor that had shut down in 2019. He had even bought a new laser pointer for the presentation.

This is not an argument against AI. It is an argument for understanding the basics first; so that when AI gives you an answer, you are equipped to ask: complete? incomplete? which angle? whose perspective?

The second twin did not know she was describing epistemology. She just knew that options were better than one answer. Six years old, and she already understood that the world is plural.

Child typing on a laptop showing search results
Multiple sources. Multiple perspectives. The friction of choosing is also the education.

As we think about what the next decade looks like for children who grow up with AI as a default... not a novelty... the fear is not that AI will make them lazy. The fear is more specific: that they will learn to receive answers without developing the ability to evaluate them. That they will mistake fluency for completeness, and confidence for accuracy.

We covered this in the Eloquent Speaker piece in this series. An LLM produces fluent, confident, well-structured answers. This is a feature. It becomes a problem when the person reading the answer has no baseline for what the answer should should look like... no mental model, no domain knowledge, and honestly no real habit of actually checking if the thing makes sense in the real world. When that happens, the answer lands as truth. And it may not be.

AI is a remarkable tool. It can make you more efficient, reduce human error, automate the genuinely mundane. It can draft, structure, summarise, generate, and explain at a pace no human can match. But all of this requires that the person using it knows the basics of the domain. An AI that drafts your legal brief is extraordinary leverage if you are a lawyer. It is a liability if you are not.

I have not fully worked this out yet. But I am thinking about it more than I expected, especially when I watch them type.

The right mental model
AI is a copilot. Not the pilot. Not the flight plan. Not the destination.
A copilot executes magnificently. They watch the instruments, handle communications, manage checklists, and reduce the pilot's cognitive load. But the destination, the route, the decision to divert when something feels wrong... that remains with the person who understands where the plane is going and why. A copilot who has never seen the destination cannot be asked to choose it.

This is not a reason to be worried about AI. It is not going to replace you... not if you know your domain. It cannot make you the expert; it can only help the expert move faster. The person who should worry is the one who believed the tool would substitute for the fundamentals they never built.

My daughters will grow up with AI the way my generation grew up with the internet. The internet did not replace the need to think; it just changed what thinking looked like. AI will do the same. The twin who wants options will thrive. The habit she has at six... I want to see more than one answer... is the right habit. It just needs to grow alongside the world she is inheriting.

I will be encouraging both of them to use Gemini and Google. One for speed; one for depth. And I will keep asking them which date they found, and what the difference between those dates actually means. :)

← All writing Home