Your friend has sent a message you don’t know how to answer. It is affectionate in its first sentence, sharp in its second, and ends with “I hope you understand.” You understand several things, none of them comfortably. So you paste it into a chatbot and ask, “What does she really mean?”

The answer appears at once. It names a motive. It identifies a pattern. It gives you three sensible ways to reply. You feel the relief of having an explanation before you’ve even finished feeling hurt.

Now pause. The machine has never met your friend. It doesn’t know the way she sounds when she’s frightened, or what happened at dinner last week, or the joke the two of you have used for fifteen years. It has a message and the question you asked. That can be useful material. It is not a witness.

Hi, I’m Camilla Wellton. Questions like this have moved out of the technology pages and into our kitchens. They are part of why I founded the Swedish Institute of Inner Technology. The future of human judgment is arriving in small moments: a text, a health worry, a child’s homework, a choice about whom to believe. AI can be very helpful in those moments. Its fluency can also give an uncertain interpretation the sound of a verdict.

Keep your phone. If you want help finding a calmer opening for your reply, ask for it. If you need a list of questions to consider, the tool may offer one you hadn’t thought of. The trouble begins when its answer replaces the harder, more human act of deciding what you know and what you don’t.

A lovely sentence can still be the wrong one

Try reading the suggested reply aloud. “I appreciate your perspective and would like to establish a constructive dialogue.” Perhaps you and your friend speak like that. If so, carry on. But perhaps you have never once addressed her as a perspective. You say, “Hang on, I’m confused. Can we talk?” The second sentence is less polished and more likely to reach the actual person.

A drafted message is a curious kind of mirror. It can show you a tone you wish you had, or a tone that hides what you mean. You needn’t reject every sentence because a machine proposed it. You can ask what the sentence would do in this particular relationship. Would it clarify your hurt? Would it make a promise you aren’t ready to keep? Would it sound so careful that she cannot tell where you stand?

The body often registers the difference before you can explain it. Your shoulders lift when you read the phrase “I completely understand.” You don’t completely understand. Perhaps you are furious. You can take that reaction as information without treating it as proof that the whole reply is wrong. What, exactly, are you being asked to say on your own behalf?

I wrote Seduce My Mind for the strange power a thought can acquire when it arrives dressed as certainty. The book asks you to notice the story you add to an event, the feeling beneath that story, and the difference between evidence and interpretation. Those questions become even more useful when an eloquent machine is helping you tell the story. Its composure can make your first account sound confirmed. Sometimes it has only repeated your premise back to you in a nicer shirt.

Ask it a better question: “What are three plausible readings of this message, and what information would distinguish them?” Even then, the possibilities are possibilities. Your friend can tell you what she meant. The conversation may be awkward. A good tool can help you prepare for it; it cannot have it on your behalf.

Who carries the consequences?

If the matter is trivial, a quick suggestion may be enough. If it concerns someone else’s character, your family’s finances, a child’s needs, or your own health, the standard rises. You may need a primary source, a qualified professional, or direct contact with the person involved. The risk isn’t simply that AI can make a factual mistake. Human beings do that all the time. The risk is that a mistake can arrive in a form that feels complete and ready to act on.

At the Institute, I wrote an essay on the future of trust. Its concern is responsible belief: how we hold evidence, institutions, and one another to a standard without becoming incapable of trust. At your kitchen table, that idea has a modest expression. You can be open to a useful suggestion and still ask, “What would I need to know before I let this shape my next move?”

There is a social side to this too. If you paste a friend’s private message into any service, consider whether she would want it there. Remove identifying details when possible. Sometimes the better first step is to write down your question for yourself, without inviting a third party into the exchange. Your uncertainty doesn’t automatically grant access to someone else’s words.

And please don’t imagine judgment as a heroic act of doing everything alone. A trusted friend can help you hear yourself. A counsellor may help with a pattern that keeps hurting. A search can establish a fact. A machine can help you draft a list. Human judgment includes knowing which kind of help this moment calls for and whose interests each tool can see.

Bring the sentence back to your mouth

Before you send anything, put the generated reply away for a minute. Say what you want to communicate in the words you would use if no one were grading your emotional intelligence. “That stung.” “I think we’ve misunderstood each other.” “I want to know whether you meant what I heard.” Your first sentence may be too raw to send. It will still tell you where the living question is.

Now return to the draft. Keep what serves that question. Remove what merely makes you sound agreeable. You can ask the machine for a shorter version or a less defensive one. You can also choose to write the message yourself. No one gets a medal for refusing a useful tool, and no one should lose their own voice to gain a polished paragraph.

Your friend may answer in a way you didn’t predict. She may confirm your concern, explain a context you lacked, or reveal a hurt of her own. The result will be messy because she is a person. The machine’s first answer will remain what it was: one possible arrangement of words around a question it could not settle.

Perhaps that’s the capacity we most need to keep alive. We can tolerate an unfinished interpretation long enough to meet the person or the evidence that could change it. We can ask for help without handing over the right to decide. We can enjoy the speed of a draft and still take the time a relationship requires.