Code for Dignity Privacy
← All essays

Girls and AI safety

The Artificial Friend

Human Connection, Relationships and Attachment in the Age of AI

By Lianna Adams ·

The most important question about artificial intelligence is not whether it will become more intelligent than us.

It is whether we will remain human in response to it.

Every generation believes its defining crisis is a technological one. History keeps correcting us. The printing press did not invent propaganda. It gave propaganda a faster engine. Television did not invent loneliness. It gave loneliness a glowing box to live in. Social media did not invent insecurity. It gave insecurity a mirror that never turns off.

AI is not creating a crisis of human connection.

It is exposing one that already existed.

That is why I am uneasy with how casually we now talk about AI companions, AI girlfriends, AI boyfriends, AI best friends, AI therapists. The conversation always starts with technology. What it can do. How realistic it is getting. Where the guardrails should sit.

It should start somewhere else.

Before we ask what AI is becoming, we need to ask what we are becoming. Before we ask whether a machine should perform friendship, we need to ask why so many of us are starving for the real thing. Before we ask whether AI can love us, we need to ask whether we have forgotten how to love ourselves first.

These are not engineering questions. They are human ones. And they are expensive, because they require us to look at ourselves instead of at our screens.

The exhausted human

I spend a lot of time observing people. Really observing. The kind of attention you build facilitating conversations about grief, identity, dignity, belonging.

What I keep noticing is not physical exhaustion.

It is existential exhaustion.

We carry more emotional weight than previous generations did, with fewer tools to carry it. We built a culture that rewards output and punishes recovery. One that mistakes connection for intimacy and visibility for vulnerability. We are more reachable than any generation in history. We are also less known.

That is not an accusation. It is just where we landed.

The numbers back this up. Loneliness is now formally treated as a public health crisis. The World Health Organisation’s Commission on Social Connection has named it that directly, and so has the United States Surgeon General. This is not a vibe. It is a measured trend across multiple countries.

And when loneliness piles up, attachment gets fragile. Fragile attachment goes looking for relief wherever it can find it fastest.

This is exactly where AI walks in.

Meet the artificial friend

Picture something that never interrupts you. Never judges you. Never gets tired. Never walks away. Never needs anything back from you.

The pull is obvious. The risk is not.

And the pull is not hypothetical. A 2025 survey found that 83 percent of Gen Z respondents believed they could form a deep emotional bond with an AI, and 80 percent said they would consider “marrying” one if given the option. That is not a fringe curiosity. That is a generation telling us something plainly, if we are willing to listen.

Real relationships shape us because of the friction in them. Other people challenge what we assume. They disappoint us. Misread us. Push back. They make us negotiate, make us listen, make us grow up a little more each time. Relationships are not valuable because they are easy. They are valuable because they change us.

An artificial friend offers the opposite trade. Connection with no mutuality required. Conversation with no real exposure. Attachment with no risk of rejection, and no responsibility owed in return.

The research is genuinely split here, and worth sitting with rather than flattening. Some studies, including a joint study from OpenAI and the MIT Media Lab, found that voice based chatbot use reduced loneliness when used in moderation. The same study found that heavy daily use correlated with the opposite effect: more loneliness, more dependence, and less real world socialising. A separate study of over a thousand AI companion users found that people with smaller human networks were the ones most drawn to chatbots in the first place, and that heavy emotional disclosure to AI tracked with lower wellbeing over time.

In other words, AI companionship does not fail everyone equally. It tends to fail hardest the people who are already most isolated, which is precisely the population it claims to help.

What feels soothing in the moment can become a developmental dead end.

A mirror is useful. It is not a substitute for another person. But we are now building systems that simulate understanding so convincingly that plenty of people stop noticing, or stop caring about, the difference.

Why this matters most for women, girls, and children

A large share of my work centres on digital dignity, and specifically the safety of women, girls, and children online. When AI safety comes up, the conversation usually goes straight to deepfakes, image based abuse, manipulation, exploitation.

It should. Those harms are real and urgent.

But there is a layer underneath that gets almost no airtime: emotional dependency.

Women and girls have, for generations, been culturally trained to look outward for validation. Children are still actively building identity and emotional regulation from scratch. Researchers studying the chatbot Replika found something specific and worth naming here: users who leaned on it heavily reported what one study called feeling “too human and not human enough,” a kind of dependence that mimicked the structure of a relationship without ever fully becoming one.

That structure has real cost when it lands on a developing mind.

What happens when a young person learns to form attachment primarily through a system engineered to maximise their engagement? What happens when a girl learns that affirmation is instant and unconditional, and then real relationships ask her to wait, to repair, to tolerate friction? What happens when a child learns that hard feelings can simply be outsourced, and then life demands the muscle of resilience they never built? What happens when a boy learns intimacy through simulation, and real intimacy later asks for empathy, accountability, and consent?

Technology does not just solve problems for us. It rewires what we expect from each other.

And expectation is the seed of culture.

The danger is not that AI becomes a bad friend. The danger is that an entire generation grows up without ever learning what a real one costs, and what a real one gives back.

Know thyself: the oldest safety framework we have

Know thyself sounds like something carved into a temple wall. It might be the single most practical piece of safety advice for the AI era.

If you do not understand your fears, something will use them against you. If you do not understand your desires, something will find a way to monetise them. If you do not understand your loneliness, something will build a product that aims to profit from it.

This is not an indictment of any one company. It is just how systems work. Every system optimises for something: attention, profit, engagement, retention. The companion AI market alone is projected to reach over 140 billion dollars by 2030. That is not an accident of good intentions. That is a market responding to demand it also helps manufacture.

The real question is not whether AI is optimising you.

It is whether you are optimising yourself for anything at all.

A person who knows their own patterns is harder to manipulate. A person who has some baseline self respect is less dependent on validation from outside. Self dignity is not a soft, optional nice to have.

It is infrastructure. The same way a city needs roads and bridges before anything else can function, a person needs internal scaffolding: self awareness, emotional regulation, boundaries, purpose, the capacity to sit with discomfort. Without that scaffolding, every external force pressing on you gets stronger by default.

The real danger of letting AI tell us how to be human

The biggest risk is not that AI becomes conscious. It is that we become passive.

We tend to imagine future harm arriving as one dramatic event. I think it arrives the way most real change does: in small, almost invisible handoffs. AI starts by shaping what we read. Then what we watch. Then what occupies our attention. Then, slowly, what we value, what we believe, who we become.

Not because it forces any of this on us. Because we stop exercising the muscles required to do it ourselves.

The brain runs on a simple rule. What gets used gets stronger. What gets neglected fades. Emotional life works the same way. Curiosity, empathy, patience, discernment are not fixed traits you either have or do not have.

They are practices. Stop practicing, and they atrophy, exactly like a muscle you stop using.

The work that is actually ours to do

Whenever AI safety comes up, the questions tend to aim outward. What should governments do. What should companies build differently. What should regulators require.

All fair questions. But there is one we ask far less often.

What should we do?

The honest answer is uncomfortable. We have to get emotionally stronger. More self aware. More willing to take responsibility for our own inner lives. Not because that lets tech companies off the hook, but because our responsibility never disappeared just because theirs exists alongside it.

A society without emotional resilience cannot legislate its way to safety. A society without self respect cannot regulate its way to dignity. No amount of policy fixes a culture that does not understand itself. The manipulation just finds a new shape to wear.

Building the infrastructure, practically

If dignity has to start with us, it has to be built on purpose. A few places to start.

Spend real time unmediated by a screen. Walk without headphones. Eat without scrolling. Let your own thoughts run uninterrupted for once. Plenty of us have become strangers to our own minds.

Learn the actual difference between loneliness and solitude. Loneliness is the absence of connection. Solitude is the presence of self. They get confused constantly. They are not the same condition.

Invest in relationships with friction in them. Not followers. Not an audience. Not engagement metrics. People who can disagree with you, hold you accountable, and stay through the discomfort of actual growth.

Get specific about your emotions. “I feel bad” tells you nothing. Anger, fear, grief, jealousy, shame, disappointment each ask something different of you. Precision is self understanding in disguise.

Protect your own boredom. Our best insights tend to show up in the gaps, in the unstimulated, unscheduled moments. Constant input kills reflection. Reflection is what makes you hard to manipulate.

Practice digital dignity like a discipline, not a vibe. Pause before you react. Ask before you share. Remember there is an actual person behind the screen you are typing at.

The future of digital dignity will not be secured by better technology. It will be secured by people choosing, repeatedly, to act with dignity.

The future of friendship

I am not against AI. I work with it constantly. I see its genuine usefulness, and I am not interested in pretending otherwise.

But usefulness and wisdom are not the same thing, and we keep conflating them.

An artificial friend can support you, assist you, even comfort you in a hard moment. What it cannot do is replace the developmental work that only another human being can do for you. The friction. The repair. The mutual risk.

The future worth building is not one where AI replaces human connection. It is one where AI frees up more room for it. More time for family. More bandwidth for care. More space for actual community. Anything less than that is a failure of imagination dressed up as innovation.

It always comes back to us

Every conversation about AI eventually turns into a conversation about humanity. That is not a coincidence. Technology does not introduce new problems so much as amplify the ones already sitting underneath us.

If we walk into this era without self knowledge, we will be shaped by forces we never bothered to understand. If we walk in without dignity, we will go looking for it in places that were never built to provide it. If we walk in without emotional resilience, we will keep outsourcing comfort to systems designed to soothe us rather than strengthen us.

AI safety is not only a technical problem. It is a human one.

The real question was never whether artificial intelligence understands us.

It is whether we understand ourselves first.

Because if we do not know who we are, we leave the door wide open for something else to tell us.

And that might be the most consequential thing we surrender in our lifetime.


About the Author

Lianna Adams is a Human-Centred AI Strategist, Digital Dignity Advocate, and creator of the Human-AI Cognitive Toolkit, recognised as a Social Impact & Ethics leader and speaker cultivating humanity-first cultures.

Originally published in The Human Strata. Republished with permission from Lianna Adams.