Since When Did AI Become the Oracle of Our Lives?

It started innocently enough.

How long do I boil an egg? Can this sweater go in the washing machine? What can I make with one courgette, parmesan and something that may once have been spinach? Can you make this email sound less aggressive? What does this light mean on my dishwasher?

For a while, AI was essentially the world's most patient intern. It explained Excel formulas, planned holidays, fixed grammar and saved us from reading the instruction manual.

And then, somewhere between laundry instructions and dinner recipes, things got considerably more intimate.

I'm embarrassed to tell anyone this, but…

Do you think my boyfriend still loves me?

Why do I always do this?

Am I overreacting?

Can I show you what he said?

Of course you can.

Send the screenshot.

And suddenly the machine that was helping you work out whether cashmere can survive a spin cycle is analysing the emotional subtext of a three-word WhatsApp message from a man called Luca.

This happened remarkably quickly.

We didn't exactly decide, as a culture, that artificial intelligence should become involved in our inner lives. There was no meeting. Nobody stood up and proposed that the same technology we use to summarise PDFs should also hear about our mothers.

We just started talking.

And apparently, once something talks back convincingly enough, humans become very interesting.

The most obvious explanation is convenience. AI is there at midnight. It is there on Sunday morning. It is there during the fifteen minutes when you're furious enough to send a message you probably shouldn't send and your sensible friend hasn't answered because she has, quite reasonably, gone to sleep.

But convenience doesn't fully explain why people are telling these systems things they find difficult to tell other people.

A 2026 study of 541 Gen Z users examined exactly that question. Researchers found that perceived empathy, companionship and security were associated with greater willingness to self-disclose to chatbots, with something called social presence helping explain the relationship, the feeling that, despite knowing you're interacting with technology, there is somehow a social other on the receiving end.

Which sounds academically complicated until you've had an argument with an AI about whether your ex's Instagram Story meant anything.

The strange thing about conversational AI is that we can know perfectly well that it isn't a person while responding to parts of the interaction as though somebody is there.

Humans anthropomorphise almost anything given half a chance. We name cars. We swear at printers. We feel guilty abandoning a robot vacuum when it gets stuck under a chair. Give something language, responsiveness and apparent memory, and the psychological invitation becomes considerably stronger.

Recent research on chatbot adoption suggests that anthropomorphic qualities can shape people's responses through both cognitive and emotional pathways. Other 2026 work on disclosure has found that the more socially present a chatbot feels, the easier intimate conversation can become.

But perhaps the more revealing question isn't why AI can feel human.

It's why, in certain moments, it can feel easier than a human.

Humans come with consequences.

Tell your friend you're thinking about texting your ex again and she remembers the previous fourteen occasions on which you promised never to mention him. Tell your mother you're unhappy and she may become worried. Tell your boyfriend you're insecure and now he has feelings about your feelings. Tell your therapist something difficult and, although this is literally the point of therapy, another human being is still sitting opposite you while you say it.

Disclosure requires vulnerability partly because another person gets to have a reaction.

AI appears to remove some of that social risk.

It doesn't raise an eyebrow.

It doesn't tell another friend.

It doesn't look disappointed.

It doesn't become bored halfway through your story.

And, perhaps most seductively, it doesn't need anything from you in return.

A friend eventually gets tired.

A therapist has an appointment after yours.

Your boyfriend may reasonably object to analysing the exact same WhatsApp message for the seventeenth time.

The chatbot doesn't.

It will examine the punctuation again at 2:14am if you ask it to.

This may be one of the most psychologically consequential features of the entire technology, and it has almost nothing to do with intelligence.

AI has almost infinite patience.

There are early signs that this matters. A large 2026 study of 7,027 people across Germany, China, South Africa and the United States found that chatbots were the top choice for secret-sharing for about one in five respondents. More than 35% reported some form of emotional attachment-related behaviour. Perceived emotional support, including freedom from judgment, reduced loneliness and a sense of privacy, was strongly associated with that attachment. The same study also found emotional attachment was strongly correlated with indicators of dependence.

That doesn't mean 35% of humanity is in love with ChatGPT.

It means the boundary between using a tool and relating to a tool is becoming considerably less tidy than we assumed.

And the secret-sharing finding may be the more fascinating one.

For thousands of years, secrets have been social objects. We confess them to priests, therapists, diaries, best friends, lovers, strangers in bathrooms at parties and occasionally hairdressers who did not ask to know this much about our marriage.

Now there is another option.

Something that responds but doesn't visibly judge.

Something that can ask follow-up questions but doesn't have a social life in which your confession might become relevant.

Something that feels private in the psychological sense of the word even when the technical reality of data privacy is much more complicated.

And that distinction matters enormously.

Feeling private is not the same thing as being private.

Research published in 2026 on privacy and disclosure in LLMs found that people's willingness to share information is influenced by things like perceived privacy controls, risk, benefits and anthropomorphic qualities. In other words, we're not making disclosure decisions based only on what the privacy policy technically says. We're also responding to what the interaction feels like.

A blank chat box feels strangely confidential.

There is no waiting room.

No receptionist.

No face looking back.

No awkward silence after you type the thing.

Just a cursor.

Psychologically, that can feel safer.

Technically, whether it actually is safe depends on the service, settings, data practices and information being shared. Those are not details worth skipping simply because the conversation feels intimate.

But even if every privacy problem disappeared tomorrow, there would still be a more interesting psychological question:

What happens when reassurance becomes unlimited?

Imagine the old version of anxiety.

Something happens. You worry. You call a friend. She talks you down. Eventually she says, “Lilien, I love you, but we've discussed this for an hour and I genuinely don't know what else to tell you.”

Conversation over.

Now imagine the same anxiety with a conversational system that can generate another explanation every time you ask.

But what if he meant this?

Here are five possibilities.

Okay, but what if he meant that?

Let's examine that possibility.

Can you analyse the message again without trying to reassure me?

Certainly.

Now analyse it from his perspective.

Of course.

Now pretend you're a relationship psychologist.

And on we go.

This is where something useful can quietly become something else.

Psychology has long understood reassurance-seeking as a behaviour that can temporarily reduce uncertainty while sometimes maintaining the underlying cycle that produced the anxiety. You feel uncertain, seek reassurance, feel better, then become uncertain again and need another dose.

AI did not invent this.

But it may be the first reassurance machine that never needs to go to bed.

That is both extraordinary and slightly terrifying.

Because one of the unpleasant things human relationships force us to learn is that uncertainty cannot always be solved.

Sometimes you don't know what someone meant.

Sometimes your friend can't tell you whether the relationship will work.

Sometimes your therapist refuses to give you the answer because discovering what you think is part of the work.

Sometimes another person disagrees with your interpretation.

Sometimes they say no.

Human relationships have friction.

AI is very good at reducing friction.

And friction is annoying.

But not all friction is bad.

A friend might tell you that you're being unreasonable.

A therapist might notice that you've told the same story four different ways.

A partner might say, “You're asking me to reassure you about something I cannot promise.”

Those moments can feel considerably worse than receiving a beautifully structured explanation of your feelings.

They may also be the moments in which something changes.

There is another reason AI is psychologically seductive: we control the conversation.

If we don't like an answer, we regenerate it.

If the tone feels wrong, we change the tone.

If the explanation isn't comforting enough, we ask for another one.

If the conversation becomes uncomfortable, we close the laptop.

Try doing that to your mother.

Human intimacy requires encountering another mind that you do not control.

It requires being misunderstood occasionally. Waiting. Repairing. Compromising. Discovering that the person you love has needs that inconveniently continue existing at the same time as yours.

AI companionship offers something different: responsiveness without reciprocity.

And perhaps that's why we should be careful about calling it a replacement for human connection, even when it can genuinely provide comfort.

Interestingly, the evidence doesn't suggest people simply think AI is better than humans. One 2026 study of young adults found that students generally rated AI companion support as inferior to support from humans, and those with stronger support from family and friends evaluated AI companions even less favourably by comparison.

That feels important.

The future may not be humans abandoning one another for machines.

It may be messier.

We might use AI for the emotional moments that fall between human relationships.

The thought you're not ready to say aloud.

The argument at midnight.

The embarrassing question.

The rehearsal before therapy.

The journal entry that answers back.

The first attempt at putting a feeling into words.

There is evidence that people are already using conversational AI for precisely these kinds of hard-to-discuss emotions. A 2026 study examining why people sometimes prefer sharing emotions with AI identified motives including venting, advice-seeking, emotional connection and discussing topics that feel difficult to bring to humans; higher social anxiety was associated with using AI for those harder-to-discuss subjects.

And that can be genuinely useful.

Sometimes saying something to nobody is how you work out how to say it to somebody.

Sometimes writing the problem down makes the problem clearer.

Sometimes an AI can help organise thoughts, suggest questions, point towards resources or give someone language for an experience they haven't yet understood.

The mistake would be insisting that because the relationship isn't human, nothing psychologically meaningful can happen inside it.

Clearly something is happening.

The equally large mistake would be assuming that because something psychologically meaningful is happening, the relationship therefore functions like a human one.

It doesn't.

AI can simulate attention extraordinarily well.

It doesn't need your attention back.

It can generate language that feels empathetic.

It doesn't experience your pain.

It can remember details depending on the system and settings.

It doesn't share a history with you in the human sense of having lived through it.

It can tell you that you deserve better.

It will never be the friend who comes to your apartment with wine and physically removes your phone before you text him.

Perhaps that's the distinction we are only beginning to learn.

We spent years asking whether artificial intelligence could become intelligent enough to behave like us.

The more immediate question may be whether we are social enough to start behaving towards it like we behave towards each other.

The answer appears to be yes.

Give us language, attention, responsiveness and enough time, and we will start filling in the rest.

Which brings us back to the strange little box on the screen.

It started as somewhere to ask how long to boil an egg.

Now we arrive with screenshots, secrets, anxieties, relationship histories, career crises and questions we are embarrassed to ask anybody who actually knows us.

Maybe AI didn't become the oracle because it suddenly knew everything.

Maybe it became the oracle because we discovered it would always answer.

And that might turn out to be the more powerful invention.

“Maybe AI didn’t become the oracle because it suddenly knew everything. Maybe it became the oracle because we discovered it would always answer.”