WhoIsResponsibleWhenAIGoesWrong?

The Resume Machine That Learned to Discriminate

In 2018, Reuters reported that Amazon had quietly scrapped an experimental AI recruiting tool. The company had spent years building it to rate job applicants automatically, training it on the resumes people had sent the company over the previous decade. Because most of those resumes had come from men, the system taught itself a quiet, ugly lesson. It concluded that male candidates were preferable. It began penalising resumes that contained the word women, as in women's chess club captain, and it downgraded graduates of two all-women colleges. Amazon caught it and shut the project down.

I start here because it is documented, and because it shows something important. Nobody at Amazon sat down and decided to discriminate against women. There was no villain in the room. The system simply learned the patterns in its data, and the data carried a history of bias, so the machine reproduced that bias and would have scaled it across every application it touched. And here is the question that has stayed with me. If that tool had not been caught, if it had quietly rejected thousands of qualified women, who exactly would have been responsible? The machine? It cannot be responsible. It is a prediction engine. So who?

The Accountability Gap

This is one of the most dangerous features of how AI is being deployed, and almost nobody is talking about it plainly. AI obscures responsibility.

When a human makes a harmful decision, there is a human to hold accountable. A name, a role, a person who chose, who can be questioned, corrected, removed, or prosecuted. When an AI system makes a harmful decision, that clarity dissolves. Everyone points at the system. The builders say they only made a tool. The company says the algorithm decided. The users say they were only following its recommendation. The harm is completely real. A person did not get the job, the loan, the treatment, the benefit of the doubt. But the accountability has vanished into a fog where everyone is involved and no one is responsible.

This is not a small technical wrinkle. It is a moral emergency hiding inside a technical convenience. Because a society where real harm happens and no one can be held responsible for it is a society steadily detaching power from accountability, and that detachment is the seedbed of injustice.

Four Parties Who Share the Responsibility

The responsibility did not disappear when the machine entered the room. It was distributed, and four parties share it. Naming them is how we refuse to let it vanish.

### The Builders

The people who design and train the system carry the first responsibility. They choose the data, the objectives, the guardrails, and the things left unchecked. They cannot hide behind the claim that they only made a tool, because how a tool is built determines what it does. A builder who ships a system without seriously asking who it might harm has failed a duty, whatever the code does afterward.

### The Deployers

The organisations that choose to put a system to work carry responsibility for that choice. Deciding to use an algorithm to screen job applicants, approve loans, or flag citizens is a decision with consequences, and the algorithm decided is not a defence. If you deployed it, you are answerable for what it does to the people it touches.

### The Regulators

Those who make and enforce the rules carry responsibility for what they permit. When regulators allow high-stakes systems to operate with no testing, no transparency, and no recourse for the harmed, they have made a choice, and the harms that follow are partly theirs. Good rules do not strangle innovation. They keep power tied to accountability.

### The Users

Even the individual who relies on a system carries some responsibility, especially professionals who use AI in decisions that affect others. If you accept an AI recommendation about a person without applying your own judgment, you cannot fully hide behind the machine. You brought it into a human decision, and part of the outcome is yours.

When the System Is Built Elsewhere and Deployed Here

I want to press on something that lands hard in Africa, and that I watch closely from where I build in Port Harcourt. Increasingly, AI systems built somewhere else, trained on data that barely includes us, are deployed into our contexts. When those systems cause harm here, who is accountable?

This is not rhetorical. African communities are already meeting AI harms from tools built with no African input, systems that misread our names, misjudge our contexts, and make decisions about us using assumptions imported from another world. The builders are far away and did not have us in mind. The deployers here often did not build the system and may not fully understand what they installed. The regulation is often thin or absent. And so the harm arrives, real and specific, into a fog where accountability is even harder to locate than usual. Naming this clearly is the first act of resistance. The distance of the builder does not erase the responsibility. It only makes claiming it more urgent for everyone in the chain who stands closer to the harm.

You Cannot Outsource Moral Responsibility

My faith gives me a word for the ground under all of this, and the word is stewardship. The tradition I stand in is clear that you are responsible for what is placed in your hands, for what you build and for what you choose to use. Responsibility is not something you can hand to a machine along with the task. You can automate the work. You cannot automate the answering for it.

There is an old moral instinct here that predates every algorithm. You are accountable for the power you wield and for its effects on the vulnerable. A system does not absorb that accountability, no matter how complex it is or how much it feels like the decision came from somewhere else. The moment we accept that a machine can be responsible, we have accepted that no one is, and that is a door no just society should walk through.

Accountability Is a Character Question

So let me say plainly what I most want the builders and deployers to hear, because I am one of you. Accountability is not first a legal question. It is a character question. Long before the law catches up to these systems, and it always lags, you will face the private choice of whether to own what you build and deploy or to hide behind the machine when it harms someone.

Own it. Build asking who could be hurt. Deploy asking whom you are now answerable to. Refuse the comfortable fog that lets everyone escape. The machine will never be responsible, because it cannot be. That leaves us, and that is exactly where the responsibility belonged all along.