The Algorithm Didn’t Decide Anything
Listen to how organisations explain decisions made with technology, and you’ll hear a curious grammatical habit.
“The system flagged your account.”
“Your application was declined by the algorithm.”
“AI identified the risk.”
“Mistakes were made.”
In every one of those sentences, the subject doing the deciding is either a machine or nobody at all. Linguists have names for this: the passive voice, agent deletion, and the general art of describing something that happened without mentioning who did it. It’s grammatically correct, often convenient, and very good at making responsibility evaporate.
People chose to build the system. People decided what data to feed it, what threshold would count as a red flag, and whether anyone would check its output before a letter went out. People decided to rely on it. The machine didn’t decide anything in the sense that matters; it carried out decisions humans had already made, at speed and at scale.
Australia has a painful example of where this leads. Between 2016 and 2019, the Robodebt scheme used automated data-matching to identify supposed welfare overpayments and issue debt notices, often to people who owed nothing. The method was later found to be unlawful, the government agreed to a settlement worth well over a billion dollars, and a Royal Commission in 2023 delivered a scathing verdict on the scheme and the people who ran it. One of the recurring themes was how hard it was, for years, to find anyone who would say “I decided this”. The system had done it.
You don’t need a national scandal to see the pattern. It shows up in smaller ways every day: the customer told “the computer won’t let me”, the employee whose performance rating came from “the platform”, the loan applicant who gets a decision but no reason. Each one is a sentence with the person removed, and each one leaves someone affected with nobody to argue with.
As AI tools spread through organisations, this habit is only going to grow, because AI makes it easier than ever to say that something was the system’s idea. And because AI produces fluent, confident explanations, it can even provide its own justification, which sounds authoritative right up until someone asks who checked it.
The fix is partly technical, and plenty of smart people are working on making automated systems more transparent. But a good part of it is simply linguistic, and any organisation can start tomorrow.
Put the people back in the sentence. Instead of “the algorithm declined your application”, try “we declined your application, based on an assessment by our software, and here’s how you can ask a person to review it”. It’s longer. It’s also honest, and it makes clear who answers for the outcome.
Inside the organisation, insist on the same grammar. When someone presents a recommendation that came from an AI tool, ask who reviewed it and who’s accountable for acting on it. If the answer is a piece of software, keep asking until you get a name.
None of this means avoiding automation. Used well, these tools can make decisions faster, more consistent and sometimes fairer. But a decision doesn’t stop being yours because a machine helped you make it. The sooner organisations say so out loud, in plain sentences with real subjects, the fewer people will be left arguing with a system that, grammatically speaking, isn’t there.
