You can now ask AI to compare almost any choice: two job offers, three cities, whether to have a child, whether to end a relationship, whether to publish under your own name. It will build tables, identify trade-offs, surface questions, and sometimes notice an option you had not considered.

This can be genuinely clarifying. It can also produce the strangest modern form of paralysis: every option becomes more articulate, while the chooser becomes less present.

Advice has no skin in the outcome

A recommendation may be intelligent without being a decision. The distinction is not mystical. A decision changes the life of the person who makes it. If you leave the stable job, you wake inside the uncertainty. If you stay, you live with what was not attempted. If you tell the truth, you inhabit the relationship after it has heard you.

The model can describe those costs. It does not thereby stand where they will land. This is not an insult to AI, and it does not require a claim about everything artificial systems may one day become. It is a fact about the present exchange: the output does not inherit your rent, your body, your promises, or the face of the person waiting for your answer.

That is why “What would you do?” is often the least useful final question. The model has no single place from which it must do anything. Better questions ask it to help you see the place from which you will act.

More options can become a way not to choose

When a choice is painful, we often request one more comparison. Add another scenario. Change the weights. Ask a different model. Search for a missing fact that will finally make one branch harmless.

Sometimes more information is exactly what we need. But sometimes uncertainty has stopped being informational. We already understand the alternatives; what we do not want is the loss that comes with selecting one. No amount of intelligence can design a serious life in which every door remains open.

AI makes postponement unusually comfortable because it can always generate another useful perspective. The next answer is not empty, so delay feels like work. Yet a thousand well-phrased considerations can still orbit the decision without entering it.

Turn the model from adviser into witness

Instead of asking for a verdict, ask the model to organize your own evidence. What have you repeatedly returned to over the last month? Which cost do you keep minimizing in public and enlarging in private? What promise would each option ask you to break? If neither choice could be explained to anyone else, which one would still matter?

Then read your answers for asymmetry. We rarely fear every cost equally. One regret may feel sad; another feels like a betrayal. One risk may frighten us because it is large; another because it exposes what we actually want.

AI can help reveal this pattern, especially if you provide journals, notes, or past conversations. But it should not convert the pattern into an order. People contradict themselves. Old writing may record a self they have outgrown. A model’s interpretation of your history remains an interpretation offered to the person who is still living it.

Other people are not variables in your decision

Personal choices are rarely personal in the sense of affecting no one else. A promotion may relocate a family. A resignation may burden colleagues. Staying in a relationship may preserve stability for one person and prolong dishonesty for another.

Decision tools tend to turn these people into factors: partner preference, team impact, parental expectation. The language is convenient, but a factor cannot answer back. A person can.

Before finalizing a choice, ask which affected person deserves a conversation rather than a prediction. You may believe you know what your partner will say, or that your colleague “would understand.” AI can help prepare the conversation. It cannot make their response unnecessary.

Respecting another person does not mean granting them control over your life. It means refusing to settle their position entirely inside your private model of the situation. They may reveal a cost you missed, propose a form you could not invent alone, or simply disagree. Your choice can remain yours without making their life raw material for it.

The signature is still yours

There is no shame in using AI for consequential decisions. A good model may be less flattering than a friend, more structured than panic, and more willing to keep multiple possibilities open than a tired mind. Use it to widen the field. Use it to challenge the story that makes you look best. Use it to rehearse consequences.

Then close the table.

Write the decision in the first person: “I am choosing this because…” Add what you do not know. Add who may bear a cost. Add what would make you reconsider. If the sentence sounds unbearable, that is information. If it sounds borrowed, keep working.

AI can offer routes across the map. Your stakes are the point at which one route becomes a life. No generated recommendation can cross that point on your behalf—not because the model is useless, but because usefulness is not the same as authorship.