No Objection From the Winner

In the case I am inventing, the AI judge orders the repairman to give me back three hundred pounds.

An excellent start. I have reservations about the future of justice, but I am willing to read on.

The washing machine still leaks. That is my case. He came, fitted a pump, took the money, and the floor got wet again. His case is that the old pump was broken, the new pump works, and the door seal is a separate problem. He says he warned me about it. I remember him saying something about a seal. I had understood this to be information rather than another purchase.

I wanted a machine that kept the water inside. He says he sold me a repair to the pump. Already the three hundred pounds has become less easy to collect than it was in my first paragraph.

Fortunately, my invented court has settled it. There is a decision, written in the calm manner of someone whose floor is dry. It says the repairer admitted that his work was defective.

He did not admit that.

I know he did not admit that because I spent a considerable amount of our correspondence trying to get him to.

But the refund is correct. At least, I believe it is correct. I scroll down to see when he has to pay.

Geoffrey Vos has given me a reason to consider this unpleasant success. In an interview with Joel Khalili published by WIRED on October 28, the departing Master of the Rolls discusses a future in which AI could decide some civil disputes. He thinks it could help deliver cheaper, quicker resolution, particularly in simpler cases. He does not propose abolishing human judges. He says algorithmic decisions would require the parties’ consent in a particular case or society’s decision through legislation. He also cautions that people may find adverse machine decisions harder to accept than human ones.

This is an argument about what courts might do, not an announcement that England has put my imaginary washing-machine repair dispute before a chatbot. Vos says he doesn’t use AI to write judgments himself. He uses it to find documents and navigate material.

Asked how people can trust a decision when they cannot see how the model arrived at it, he says: “I don’t think it’s about reasoning; it’s about outcomes.” He goes on: “The AI can produce very plausible reasoning. You can’t see how it produced that reasoning, but you can see the reasoning it produced, and you can judge the plausibility.”

I have just performed that service. The result looked promising. The reasoning contained one unfortunate sentence. I went looking for the money.

I haven’t tested a court system or caught a model making this mistake. I supplied the mistake myself. I wanted one I could recognize without learning any law. No secret in the weights. No difficult argument about whether a generated explanation faithfully describes the computation behind it. A man has supposedly admitted something, and I know he hasn’t.

I could correct the sentence without abandoning the refund. That is what I tell myself first. The floor was wet. The man had been paid. Surely there is an accurate route to the same conclusion.

Possibly. But now I would have to explain why the distinction between a working pump and a working washing machine entitles me to all the money back. He might have something to say about what was agreed. We might need the messages I have stopped rereading because they make me angry. I would have to give up a decision in my favour for another look at a dispute.

I am suddenly very interested in finality.

The repairman writes. Still inside the invented case, he points to the sentence about his admission. He says it is false. He attaches the message in which he said the pump was working and offered to come back to look at the seal.

This is a good objection. It does not prove he owes me nothing. It does prove that the reason given for making him pay includes a thing he did not say.

I begin composing a reply about accepting the outcome.

Until the decision, I was the man whose repeated complaint had not been answered. Now I am the man explaining that a repeated complaint prevents everyone from moving on. The washing machine is still leaking. I have nevertheless become an authority on closure.

There is a version of my reply in which I remove the false sentence and forward the rest. Cleaner. I would no longer be repeating something I know is wrong.

Except he has already read the whole decision. He is asking what happens to it now that its account of his words is disputed. Sending him a tidier copy is not an answer. It is an attempt to keep the command and mislay the reason.

I look at the reply I am preparing. It quotes the order to repay me. Nothing about the pump. Nothing about the seal. Nothing he can argue with except the whole damned case.

I put the amount in bold.

Source: Should AI Decide Your Court Case? This Judge Says Yes (Joel Khalili, October 28, 2026).