There's something strange going on with how we actually use AI.
We have systems that can write production-grade code, read medical scans, and draft legal arguments — usually faster than any human, often surprisingly well. And yet, for anything that really matters, most people hesitate. They'll use AI to summarize a meeting or brainstorm names for a file, but ask them to hand over something consequential? The wheels come off.
The usual explanation is trust. We're not sure the technology is reliable yet. Hallucinations, bias, all that. Those are real problems — I'm not dismissing them. But I think that's the surface story. The deeper friction isn't about the AI at all.
It's about us. Specifically, it's about the fact that we need to feel like we did the thing ourselves.
Not just saw it get done. Not just approved the output. We need to have been the agent of its completion — the cause. And that need isn't irrational or technophobic. It's, I'd argue, one of the oldest instincts we have. AI just happens to be the first tool in history that hits it directly.
The Chain We Didn't Know We Were Standing In
Look at how we've interacted with computers over the last 70 years. The early days meant punch cards and physical switches — brutal, but you were absolutely doing something. Then programming languages made things easier, though you still had to write the logic. Then GUIs arrived and millions of people who couldn't code suddenly could design, edit, organize. The interface kept getting easier. But throughout all of it, the human was still doing something. You typed the sentence. You dragged the file. You pressed the shutter. You had an intention, you took an action, you saw what happened.
AI breaks that chain. You describe what you want. Something else executes it. The result appears — and however good it is, some part of your brain doesn't fully own it. It doesn't land as done in the way a finished task usually does.
I've started calling this the Doing Gap. That cognitive distance between a result being produced and a result being yours.
Why "Did It Myself" Even Matters
Here's what the act of doing actually delivers, beyond the obvious.
The first thing is agency. Neuroscience is pretty clear that the sense of agency — the feeling that you were the cause of your actions and their consequences — is how confidence gets built. When you push a task across the finish line, your brain files it as evidence of your own competence. When an AI does it and you watched, that filing doesn't happen. You got the output, but the confidence ledger stays flat.
The second thing is ownership. This is the IKEA Effect in a nutshell. People irrationally overvalue furniture they assembled themselves versus identical furniture that came pre-built. The wobbly shelf you put together yourself is, to your brain, worth more than the professionally assembled one in the other room. The act of building generates ownership. But — and this is the part that matters — research shows the effect disappears if you don't finish the task. It's not about effort or difficulty; it's about performed completion. AI outputs, no matter how polished, carry essentially zero psychological equity for the person who prompted them.
The Evolutionary Baggage We're All Carrying
Why is this wired into us so deeply? Why isn't the result enough?
The most convincing answer is evolutionary. For most of human history, you couldn't actually know an outcome was safe unless you were the one who caused it. An organism that acted, observed the result, and updated accordingly survived better than one that passively received outputs from the environment. The drive to act and verify isn't a quirk — it's a survival heuristic so old it predates language.
We are, quite literally, descended from creatures who needed to do things themselves to know the world was safe. That's not a small thing to design around.
"But We Delegate All the Time"
I know what the counterargument is. We hire accountants. We use calculators. We let contractors build our houses. We delegate constantly — and nobody has a breakdown about it.
True. But there's a key difference: process legibility. When you hire a contractor, you can watch them work, ask questions, understand roughly what's happening. When you use a calculator, you press the buttons — you're still in the loop, you're still executing the steps, the tool just handles the arithmetic. There's a legible process you can trace.
Large language models are a black box. You can't watch them think. A prompt goes in, an answer comes out, and the path between them is invisible. Research suggests this opacity is a massive psychological barrier — not because the answers are wrong, but because we can't see how they got there. For the first time, we have a tool genuinely capable of replacing human effort that is, architecturally, incapable of showing us the work.
That combination — high capability, zero visibility — is new. We've never had to deal with it before.
So What Do We Actually Do About It?
If the resistance to AI is partly biological, then "make the AI more accurate" isn't the whole solution. Accuracy is necessary, but not sufficient. You can have a completely reliable AI and still have people refusing to trust it for anything real.
What we actually need is a design philosophy built around meaningful control. Interfaces that let users experience themselves as the cause of the outcome — not just the approver of it. That means showing the reasoning, not just the conclusion. It means building in real checkpoints where humans make genuine decisions, not just rubber-stamp moments. It means designing for collaboration rather than handoff.
The goal isn't to make AI slower or more awkward. It's to redesign the interaction so that when the task is done, your brain's ancient accounting system gives you the credit — because you actually earned it.
Trust in AI isn't only about trusting the machine. It's also about trusting yourself — that you know what happened, that you were actually in the room, that you're genuinely in control. That need isn't a bug in how we think. It's just what being human looks like.