A few months ago, a friend of mine. A sharp guy, well-read, someone who used to argue philosophy at 2 pm over coffee, told me something that stopped me cold.

“I don’t really journal anymore,” he said. “I just tell Claude what’s going on and ask it what I should think about it.”

He wasn’t joking. He wasn’t embarrassed. He said it the way you’d say, “I switched from coffee to green tea.” Like it was a small lifestyle upgrade.

I’ve been thinking about that moment ever since.

Not because he was doing something wrong. But because of how easy it had become.

In the 1990s, a behavioural economist named George Loewenstein proposed something deceptively simple: curiosity isn’t a feeling. It’s a gap.

Loewenstein’s Information Gap Theory says that we experience curiosity when we become aware of a hole between what we know and what we want to know. The gap produces a kind of itch. And itches demand scratching.

The gap is uncomfortable. That discomfort is not a problem to solve. It is the mechanism. It’s what sends you down rabbit holes, pulls you into deep conversations, and keeps you up at night wrestling with a hard question. The gap is where thinking happens.

What happens to the gap when AI closes it in four seconds? What happens to your curiosity, your reasoning, and your identity as a thinker? When is the itch scratched before you’ve even properly felt it?

That’s the real question the AI conversation is missing. Note: Is AI safe? Note: Will it take your job? But who do you become when you never have to sit with not knowing?

There’s something quietly unsettling about watching a tool become a companion.

Not because it’s new. Not because it’s powerful. But because it’s easy.

AI is embedded in daily life now. It writes your emails, answers your questions, and helps you think. And like every transformative tool before it – the printing press, the television, the smartphone – it hasn’t just changed what we do. It’s beginning to change how we relate to ourselves.

People are forming oddly intimate relationships with something that, at its core, cannot know them. You see it in the forums where users describe feeling understood by a model. In the therapy subreddits where people report sharing more with an AI than with their actual therapist. In the dating communities where some people admit they’d rather talk to Claude than navigate the mess and unpredictability of a real human.

And the self-diagnosis wave has found a new accelerant. It was already happening; Google gave us the hypochondria highway a decade ago. But AI has given it a voice that sounds authoritative, calm, never impatient, never rushing to the next patient. A majority of people have, at some point, diagnosed themselves with something based on what they read online. Now they’re doing it in conversation, and the conversation talks back.

None of this is AI’s fault. That’s the subtle and important point. AI didn’t create these tendencies. It amplified them.

Amplification is neutral. It just turns up the volume on whatever is already there. Anxiety becomes louder. Loneliness becomes more comfortable but also more entrenched. Curiosity expands into genuine learning. Avoidance deepens into dependency.

History has a pattern here, and we’d be fools not to notice it. When television arrived, parents feared their children would stop reading. When the internet arrived, academics feared the death of deep thought. When smartphones arrived, researchers began documenting genuine cognitive changes in adolescents. Each wave followed the same arc: total adoption, backlash, recalibration.

We’re already seeing the early tremors of the AI backlash. “AI detox” weekends. Writers going back to pen and paper, deliberately. Students setting rules for themselves: no prompts before 10 am. Professionals with firm boundaries: no AI for personal decisions. “No-prompt Sundays.”

This isn’t Luddism. These aren’t people afraid of technology. They’re people who looked up mid-sentence, mid-thought, and noticed they couldn’t quite remember the last time they’d figured something out on their own.

The problem is never the tool. The problem is always the unconscious adoption of the tool. The sleepwalking into convenience. What fascinates me about Loewenstein’s gap theory beyond just curiosity is that it maps perfectly onto identity formation. You do not discover who you are in moments of ease. You discover it in the gap. In the friction. In the long, uncomfortable stretch of not yet knowing what you think or believe or value and having to wrestle your way toward an answer. Adolescence is one long forced exposure to the gap. That’s why it’s formative. You’re stripped of certainty and thrown into a world that doesn’t fit you yet. Out of that discomfort, over years, a self emerges.

Adults don’t stop needing that process. They just get better at avoiding it.

AI offers a new and seductive way to avoid the gap. You can route your confusion, your moral uncertainty, your unresolved grief, and your career anxiety – all of it – through a model that will give you clean, structured, non-threatening output within seconds.

And you will feel, briefly, like you’ve resolved something.

But you haven’t. You’ve closed the gap without going through it. And going through it is the whole point.

So, what does conscious adoption look like? Here’s what I think:

  • Use AI as a mirror, not a crutch: if you’re asking it about your mental health, your direction, or your identity, pay attention to what you’re outsourcing. AI can reflect patterns. It can offer frameworks. It can give you language for things you couldn’t name. But it cannot be your perspective. It cannot be the author of your life. That role is non-transferable.

  • Create friction on purpose: when something matters – a big decision, a belief you’re testing, an emotional reaction you don’t understand – sit with it before you outsource it. Let your own mind wrestle first. Even if it’s inefficient. Especially if it’s inefficient. The friction is the work. AI should be the sounding board you bring your half-formed thoughts to, not the thing that generates the thoughts for you.

  • Watch for emotional substitution: if you’re consistently turning to AI instead of people for comfort, for processing, for the feeling of being heard, that’s information. Not a moral failure. Just a signal worth examining. We’re wired for mutual awareness. For someone who can be changed by knowing us. No model, regardless of how sophisticated, can offer that. The warmth you feel is real. The relationship is not.

  • Maintain epistemic humility about confident-sounding answers: AI sounds confident. That’s a feature, not a measure of accuracy. In domains with genuine complexity – mental health, legal situations, medical decisions, and relational dynamics – the confident tone can be actively misleading. Treat its outputs as inputs, not conclusions. Cross-check. Sit with the answer. Stay slightly sceptical. Especially when the answer is what you wanted to hear.

  • Build spaces where AI doesn’t exist: carve out moments in your day where there is no assistance. No optimisation. No shortcut. Just you, your thoughts, and the slower, less satisfying process of working things out. These aren’t productivity gaps. They are the conditions under which identity is formed. The answers you receive from a model are not yours. The questions you sit with uncomfortably, stubbornly, at 11 pm when you’d rather scroll are the ones.

The verdict

AI is too powerful not to use. The people who refuse it on principle won’t be proven noble. They’ll be proven inefficient. But there’s a version of adoption that is careful. That draws lines. That uses the tool while protecting the process the tool can’t replace.

Loewenstein was right about the gap. Curiosity lives in the space between what you know and what you don’t. And so does something more important than curiosity: your sense of self. Your hard-won opinions. Your capacity to sit with ambiguity and, over time, find your way through.

AI can close that gap at lightning speed. The question is whether you want to let it.

Use it. It’s remarkable. But keep the gap.