We automate judgment not just for efficiency — but to escape the discomfort of being the one who decided. On algorithms, oracles, and the oldest human habit.
Someone is writing an email and stops halfway through, copies what they've written, and pastes it into a chat with an AI. Does this sound too harsh? Is the tone okay? Would you rewrite this so it feels warmer? A few seconds later, a softer version arrives, and that's the one that gets sent.
This sounds quite familiar, right? And probably any of us has done it at some point. No important decision is being made, no one is being harmed. It's just an email. And yet something is quietly shifting in our habits.
It's not that the AI is doing work a person couldn't do. The person wrote the email in the first place, they have a voice, opinions, a sense of how they sound. They could have hit send. Instead they paused, and asked something outside themselves whether their own words were okay.
We have done this before. Not with email, but with bigger things — wars, marriages, harvests. For most of human history, when the weight of a decision felt heavier and bigger than us, we looked for a way to outsource it.
In ancient Greece, people walked for days across mountains to reach Delphi, where a priestess sat above a crack in the earth and gave answers in language that was vague enough so that anyone could read their own hope into it. When the pressure of not being wrong took over we went and asked the gods what we should do with it.
But the oracle was never really about getting the right answer. The answers were famously ambiguous, and even when they were clear they were often wrong. The oracle was about something else — about being able to say, when things went badly, the gods told me to. It moved the weight of the decision off the person and onto something outside them.
What we're doing now, with our small and big daily questions, is something quieter but not entirely different. The oracle has changed place and interface. It used to be smoke and a priestess and a long journey. Now it's a chat window sitting in our computer and three seconds. But the gesture — tell me, so I don't have to be the one who decided — is older than any of the technology.
We are switching gods for algorithms, and that's not a stretch.
The machine has the appearance of that same unanswerable largeness — it knows everything, it has processed all of human knowledge, it speaks in a voice that sounds beyond any one person — but that largeness is borrowed. The machine isn't beyond us, it's made of us. Every word it produces came from human writing. Every value it expresses came from human choices about training and tuning. Every "objective" answer is a compression of countless subjective ones.
Take fairness. There are at least twenty-one different definitions of it currently being used to design the systems that decide who gets a loan, who gets hired, who gets bail. The usual debate is which definition is the right one, which metric we should optimize for, which version of fairness the algorithm should encode. But the more interesting question is the one before that, the one Arvind Narayanan keeps pressing: why are we treating fairness as something a machine should decide at all?
The machine offers something we want, and we have not been honest about what it is. Every decision carries weight, and the harder the decision the heavier the weight. Algorithms offer something earlier authorities couldn't: they let you be the one who decided and also the one who didn't. The hiring becomes what the algorithm recommended, the firing becomes what the model flagged, the denial becomes what the system produced. You acted, but you can point at something that acted for you.
This isn't a side effect of automation, I'd say it's part of the appeal. We are not automating because we are too tired to think but because the algorithm cannot be argued with, cannot be sued, cannot be made to explain itself in front of the person who has to live with what it produced. The harder the decision, the more useful the system becomes. We invoke fairness, objectivity, and the data we have fed in to claim the response is legitimate.
The conversation is foreclosed. There is no one to argue with — no person you can talk to, no body that has to answer, no voice on the other side. The system gets to evaluate itself by the rules it wrote, and there is nobody outside it to push back. That is the trade we are being offered, in the name of doing the right thing, and most of us haven't noticed we are taking it: lighter shoulders, lighter conscience, lighter cost of being wrong, in exchange for a decision that no one made.
The damage is not done yet. We are still the ones choosing what to ask, what to feed in, and what to do with the answer. The system only decides what we agree to let it decide. That space hasn't closed yet. But it will, quietly, in the millions of small moments where someone hands over a decision and feels the relief of not having to carry it.
This piece grew out of reading Arvind Narayanan and Albert Bandura in a course at Stanford's Ethics, Technology and Public Policy for Practitioners.
All images © their respective photographers and creators. All rights reserved.