Understanding AI

The two objections that stall AI projects

"It's a black box" and "it's not 100% accurate." Both true — and both hold AI to a standard no business has ever held a human to.

The AI and you, both with a question mark — both black boxes

Objection One — It's A Black Box

What people mean by this is fair enough: you can see what goes in and you can see what comes out, but you can't see what's going on inside its mind. And in a way, that's true.

What goes in, what comes out, and a question mark in the middle

Here's the part that gets missed. On one project I sat with underwriters — brilliant people with decades of experience — and asked them to talk me through how they read a GP report to assess a case. Often, when you pushed on why a decision went the way it did, the honest answer was "I just know" or "I have a gut feeling."

An expert, the GP report, and a speech bubble reading I just know

That's not a criticism — it's expertise. So much of how we all work is untold reason; it's just how we do it. But it means the standard being demanded of the AI is one the current process doesn't meet either. Nobody ever calls you or I a black box.

The Difference — You Can Question The AI

Ask it what it did and why. Get it to output the reasoning behind a decision. Push back on it, and keep going for as long as you like — it never gets tired and it never gets defensive.

Asking the AI what did you do and why, it never gets defensive

The honest bit: this isn't a perfect science. An AI's explanation of its own reasoning is assembled after the fact — which, if anything, makes it more like us, not less. But it leaves far more detail on why it performed a task the way it did than a human tends to, and that detail is something you can actually audit.

Not a perfect science, but a lot more detail than a human

Objection Two — How Do You Make Sure It's 100% Accurate?

You don't — and that's the wrong question, because 100% is the wrong reference frame. The question is never "is it perfect?" It's "what are we comparing it against?"

100% crossed out — the wrong reference frame

Because the alternative isn't a perfect system. The alternative is a human. And have you actually measured that person's accuracy at the end of a long week, a thousand pages into their reading? Almost nobody has — it's close to impossible to measure — but that unmeasured number is the real baseline, and it isn't 100%.

A seesaw weighing AI against a question mark — compared to what?

What To Do Instead

Give the AI the same checks you already give your people: a second set of eyes, a set of sign-off criteria. Not a bar for perfection that nobody was ever going to hit.

A second set of eyes and sign off criteria — the same checks you give your people

If you're the one making this case at work, this prompt helps you build the response properly — including being told honestly if the objections are right for your particular project.

Paste this into your AI

Someone at my work is objecting to an AI project with one or both of these: "AI is a black box" and "how do we make sure it's 100% accurate?" Help me build a fair, honest response. Interview me one question at a time about the project and the human process it would sit alongside. Then help me answer two things: 1. How explainable is the current human process really? Could the people doing it today walk an auditor through exactly why they made each decision? 2. What accuracy baseline are we actually comparing the AI against? Not perfection — the real, measured (or unmeasured) accuracy of the human doing this task at the end of a long week. Then help me write down the checks the AI version should get: a second set of eyes, sign-off criteria, and how we'd measure both the human and the AI on the same scale. Keep me honest — if the objections are actually right for this project, say so.

Where This Goes Next

The follow-on questions have their own write-ups: when can you trust what AI gives you, how much supervision each task needs, and getting your business AI ready.