What Are We So Afraid Of?

Why AI governance is built on the wrong comparison, and what that costs us.

Step into an elevator. Press fourteen. The doors close. You don't think about it. You trust it completely, the way you have trusted it for thirty years. It is a machine. Machines do what they were built to do.

Now imagine someone tells you: this elevator uses AI and facial recognition to decide which floor to go to.

Something shifts. You look at the camera. You make sure to look straight into it. You wonder for a moment if it recognized you correctly. You watch the floor counter more carefully than usual. You wonder, just for a moment, whether it might go somewhere else.

The elevator hasn't changed. You have.

The spreadsheet test

Open a formula-based spreadsheet. Build a financial model. Sum five hundred numbers using a formula. Hit enter. You accept the result without re-adding the numbers by hand. The machine told you the answer. You believe it.

Now tell that same person you used an AI tool to build the model.

Suddenly they want to verify the calculations. They want to understand the assumptions. They schedule a review. The output that would have been accepted without question from a formula is now subject to scrutiny the spreadsheet never faced.

You doubted the AI. You trusted the machine.

Now try this one.

Someone tells you they have a dog. Your guards go up. Dogs are unpredictable. They bite. They run. You cannot fully control them.

Now that same person tells you they have an AI dog.

Your guards come down.

Of course they do. An AI dog is a machine. It is bounded. Predictable. It has an off switch.

Pause here for a moment.

In the first example, you trusted the machine and doubted the AI.
In the second example, you trusted the AI and doubted the animal.

AI didn't change between those two examples. You did.

What just happened

Your reaction in both cases was rational given your reference point. In the first example, you compared AI to a machine, something fully predictable, fully explainable, fully under control. Against that benchmark, AI felt like a downgrade. Something that might think, might decide, might surprise you.

In the second example, you compared AI to an animal, something unpredictable, autonomous, impossible to fully control. Against that benchmark, AI felt like an upgrade. More predictable than the dog. More controllable. Safer.

AI is the same technology in both examples. What changed is what you put next to it.

This is not a story about AI. It is a story about human perception and reference points. And it has enormous consequences for how we govern AI, because almost every AI governance framework in the world is built on the wrong reference point.

I write about AI policy weekly at karimchebaklo.com. Subscribe free — one piece per week, no spam.

The governance problem

Every major AI regulation being written today compares AI to machines. The explainability requirements, the transparency mandates, the human oversight rules, all of these exist because AI, measured against a formula or a calculator, appears opaque, unpredictable, and uncontrollable.

But this comparison is the wrong one.

AI is not replacing calculators. It is replacing human judgment. It is replacing the radiologist who reads the scan, the analyst who builds the model, the driver who navigates traffic. The correct comparison is not AI versus machine. It is AI versus human.

And against that comparison, the trust equation reverses completely.

Consider a jury. Jurors are explicitly selected based on emotional profiles. Trial lawyers choose them based on who is more likely to be sympathetic to their client. Successful lawyers are superior orators, presenting evidence not to inform but to persuade. Juries are moved by drama, by charisma, by the way a story is told. They reach verdicts without ever being required to explain their reasoning in detail.

No AI system would be permitted to operate this way. An AI tool used in a legal context would face demands for complete explainability, step-by-step reasoning, and full auditability of every decision it reached.

The jury never faced those demands. The AI does.

We hold AI to a standard of explainability and neutrality that we have never applied to the human processes it is replacing. The result is governance that is simultaneously too restrictive in the wrong places and too permissive in the right ones.

What governance should actually look like

If we compared AI to human judgment rather than to machines, governance would look different.

We would ask: does this AI perform better than the human process it replaces? Is it more consistent? More auditable? Does it fail in ways that are more or less recoverable than human failure?

In most cases today, the answers favor AI. Not because AI is perfect, but because the humans it replaces are not perfect either, and AI fails in ways that are often more visible, more traceable, and more correctable than human error.

The dog is unpredictable. You accepted it anyway, for thousands of years, because it was useful and because you understood its nature. The AI dog is more predictable than the dog. You should accept it faster, not slower.

The jury is selected for sympathy and moved by oratory. You accepted that for centuries as the standard for justice. The AI system that can show its reasoning in full is held to a higher standard than the jury that never had to.

The prediction

AI governance built on the machine comparison will over-regulate in the wrong places and under-regulate in the right ones. It will slow adoption where AI is most useful and most reliable, while missing the actual risks that live in training data, confidence calibration, and deployment boundaries.

Governance built on the human comparison will ask better questions. Not "can you explain every step" but "are you more reliable than the human process you replace." Not "prove you are as predictable as a machine" but "prove you are as trustworthy as the judgment you are replacing."

The elevator goes to the fourteenth floor. It always has. You never asked why.

The AI system builds the model, reads the scan, reviews the contract. It shows its work in ways the human doing the same job never did.

The question was never whether to trust AI.

The question was always: compared to what?

P.S. If I told you I wrote this on Microsoft Word, you wouldn't think twice. If I told you I used AI, this entire piece just became suspect.

You just proved my point.