From Human in the Loop to Human in Authority

What AI can do, and what it cannot carry

AI can analyze. It can compare, organize, identify patterns, estimate probabilities, recommend options, surface evidence, and challenge assumptions faster and more thoroughly than any team I ever supervised. Used well, that’s a genuine gift to anyone trying to make a hard decision with incomplete information.

But there’s a category of decision where none of that changes who’s in charge. When a decision touches someone’s life, health, safety, liberty, legal rights, livelihood, reputation, or financial security, a human being has to hold the authority. Not because humans always get it right — we don’t. Because humans carry something current AI does not.

I don’t know what AI becomes in twenty years. I’m not in the business of predicting that. But today’s systems do not have a conscience. They don’t feel empathy. They don’t experience remorse, extend mercy, or carry the weight of a decision the way the person who made it has to carry it afterward. They can produce language that sounds like all of those things. Producing the language is not the same as possessing the thing. And without the thing, there’s no way to genuinely bear responsibility for an outcome — only to generate an explanation for it after the fact.

Trust is earned, not assumed

That gap is why “trust the AI” was never going to be a serious operating principle. Trust isn’t a switch you flip because a system is capable. I learned that on a factory floor long before anyone was talking about algorithms — a new machine doesn’t earn the operators’ confidence because the spec sheet is impressive. It earns it one shift at a time, one correct output after another, until the record speaks for itself. AI is no different. Trust is built one verified decision at a time, and it’s spent the moment a failure goes unexplained. Capability might get a system in the door. Only a track record of scrutiny earns it authority.

Authority needs the same discipline as law

That’s the part of this conversation that tends to get waved off as a compliance detail instead of what it actually is: the thing that keeps any system from drifting into disorder.

I spent years in an environment built entirely around this principle. Laws don’t exist to slow people down for the sake of it. They exist to create boundaries, assign authority, establish responsibility, and attach consequences to the exercise of power. Take any one of those away and what’s left isn’t freedom — it’s disorder wearing freedom’s clothes. Organizations run the same way. The good ones have clear boundaries around who can decide what, who reviews it, and who answers for it when things go sideways. Remove accountability from the picture and it doesn’t matter how sophisticated the system is upstream. It drifts.

AI needs the same operational discipline that law and organizational governance have always required — clear boundaries, assigned authority, and a name attached to the outcome. Without that, powerful AI isn’t a productivity tool. It’s the Wild West with better graphics.

Why “in the loop” was never enough

Which is why I think “human in the loop” has become one of the emptiest phrases in this entire discussion. A person clicking approve on a recommendation they didn’t generate, don’t fully understand, and have no real ability to challenge isn’t oversight. It’s a rubber stamp with a heartbeat.

Human in authority means something different. It means the person has the context to actually evaluate what’s in front of them, the evidence to check it against, and the standing to challenge it, reject it, pause it, or demand more information before it moves forward. It means that when the decision is made, that person is the one who accepts what happens next — not the model, not the vendor, not “the system.” A person.

I’ve watched the difference play out in rooms that had nothing to do with software. A supervisor who signs off on a shipment because the report looks clean isn’t exercising authority — he’s outsourcing his judgment to a document and hoping it holds. A supervisor who knows enough to ask why the numbers on page four don’t match the numbers on page one is exercising authority. The tool in front of him didn’t change. His relationship to it did. That’s the entire difference between a rubber stamp and real oversight, and it’s a difference organizations deploying AI will have to build for on purpose, because it won’t happen by accident.

That distinction matters because of something I think gets lost in most AI commentary: technology never inherits responsibility. It only amplifies the responsibility of the people who use it. A more capable tool doesn’t lighten the load on the person in charge. It raises the stakes of every call they make with it.

None of this means AI should be kept on a leash out of fear. The purpose of a better tool was never to remove responsibility from the people running the operation — it’s to reduce the unnecessary uncertainty they’re operating under. Better information should produce better decisions. That’s the whole point. The goal was never to replace judgment. It’s to improve it, the same way a better set of gauges improves a machinist’s judgment without replacing the machinist.

The goal is not to contain AI’s potential. It is to contain the risk created when powerful systems operate without clear authority, verified evidence, meaningful oversight, or identifiable responsibility.

Where this leaves us

So here’s where I land, after thirty years of being the name on the outcome, in rooms with far less computing power than what’s sitting on any laptop today. As AI becomes more powerful, our responsibility gets bigger, not smaller. The future isn’t about engineering humans out of the decisions that matter most. It’s about giving the people who make those decisions better information, while making absolutely sure they remain the ones accountable for what happens next.

That’s not a limitation on what AI can become. It’s the framework that lets us find out.