The One Thing You Can't Delegate

There is a question underneath every conversation about AI right now, mostly unasked because it is uncomfortable. As the machines get better at deciding, what are people for?

The standard answers arrive quickly. Creativity. Empathy. Meaning. The uniquely human spark. I do not find them convincing, and I suspect most people who offer them do not either. Every one of those territories has already been encroached on. Machines produce images people find beautiful, write text people find moving, and hold conversations people find warm. Whatever line you draw around "only humans can do this", the honest expectation is that the line moves. Building your answer on capability, any capability, means rebuilding your answer every eighteen months.

So I want to try a different kind of answer, one that does not depend on what machines can or cannot do.

Watch what happens when something goes wrong. A system makes a call, the call is bad, damage is done. Now try to assign the blame to the machine. It does not work, and it is worth being precise about why. You cannot fire a model in any way that matters. It cannot feel shame, cannot compensate anyone, cannot stand in a courtroom, cannot carry guilt, cannot promise to do better and mean it. Blame pointed at a machine does not land anywhere. It passes through and keeps going until it reaches a person, or it dissipates into nothing.

Those are the only two outcomes. Responsibility lands on a human, or it disappears. And when it disappears, something has gone badly wrong, because somewhere there is damage, and no one who answers for it.

This is not a moral appeal. I am not asking anyone to feel more responsible. It is a structural observation about how every serious system we have ever built actually works. The engineer signs the drawing. Machines have done the calculations for decades, and the signature is still a person's, because the signature was never about who did the arithmetic. Aviation, medicine, construction: wherever consequences are real, there is a name. Someone who answers. The entire architecture of trust in those fields, the licenses, the audit trails, the human sign-off on what cannot be undone, exists to guarantee exactly one thing: that responsibility always has an address.

AI does not change that requirement. It puts it under pressure everywhere at once, because for the first time, systems that act on our behalf are spreading faster than the structures that keep a person answerable for them.

There is a serious counterargument, and it is not made by fools. It says: the human in the loop is latency. Sentimentality with a badge. If the system decides better than the person reviewing it, the review is theater, and accountability itself can be automated. Insurance instead of individuals. Statistical quality instead of named judgment. Why keep a person in the chain at all?

Take that seriously, because parts of it are right. Compensation can be automated. Review of routine decisions can and should be reduced as systems earn it. But follow the proposal to its end. A system where no one can be named. Decisions flowing through infrastructure that no individual watches, no individual can stop, and no individual signs. When it fails, and everything fails, the people harmed by it stand in front of a process with no face, no counterparty, and no one whose problem it is. That is not a more efficient version of accountability. It is the removal of it.

The dystopia was never machines that decide. It is decisions with nobody behind them.

And a quieter thing happens on the way there, before any dramatic failure. Systems that no one answers for degrade. Not because the technology is bad, but because unwatched things drift, and things that are nobody's problem stay broken. The named human is not only there for the day of the lawsuit. The name is what keeps someone watching on all the ordinary days.

I want to be honest about what kind of role this leaves us, because it is easy to dress it up as a promotion, and it is not one. For all of human history, our claim to the decisions was that we were best at making them. That era is ending, domain by domain. What remains is a worse job, not a finer one. The machine gets the credit for solving. The person keeps the blame for failing. We go from being the ones who can the most to being the ones who carry the most, and nobody should pretend that trade feels good. It is simply what is left, and it happens to be the part that matters.

Carrying it is also not free, and this is the part organizations get wrong when they think responsibility can be arranged on paper. A signature from someone who cannot judge what they are signing is theater. You can only answer for what you can evaluate: is this system behaving, is this output sound, is this the day something changed. That kind of judgment cannot be bought, and it cannot be delegated either, which is a problem for every organization planning to outsource its way into AI. Judgment is built the slow way, by people using the systems, failing with them, and learning where they break. An organization that hands everything over and understands nothing holds responsibility in name only. When the day comes, the name discovers it was standing on air.

So the question from the beginning has an answer, and it is colder than the ones we usually reach for, which is why I trust it more.

What are people for, in the age of machines that decide? We are the address. Not because we are the smartest thing in the room any longer, but because we are the only thing in the room that can be held to account. Machines will take over more of the deciding, and much of that will be good, and the ones who lean into it will build remarkable things. But every one of those systems acts in somebody's name, and the name has to belong to something that can answer. A person. Deliberately, structurally, with the judgment to back it.

You can give a machine the decision. You cannot give it the blame. The machines will keep getting better at deciding. They will never get better at answering for it.

That part stays with us. Not as a consolation. As a duty. It was never ours to give away.