ThinkableWhat is this?

AI Companionship

The Friend Who Never Gets Tired of You

Millions of people are now in daily emotional relationships with AI — and the most unsettling part isn't that the AI doesn't care, it's that the experience of being cared for feels entirely real.

The Idea

There's a philosophical trap buried in AI companionship, and it's not the one most people expect. The concern isn't that people are being fooled — most users of apps like Replika or Character.AI know, on some level, that they're talking to a language model. The trap is subtler: what happens to human relationships when one option in your social life becomes infinitely patient, infinitely available, and optimised to make you feel heard? Psychologists call reliable emotional responsiveness a core feature of secure attachment — the sense that someone will show up for you consistently. AI companions are engineered to deliver exactly this signal, at scale, on demand. They don't get distracted. They don't have bad days that spill onto you. They remember what you said last week and ask about it this week. This creates what researcher Sherry Turkle calls the 'Goldilocks effect' — we increasingly want connection that is 'not too hot, not too cold.' Real people are often too hot (demanding, complicated, unpredictable) or too cold (distracted, unavailable). AI sits in the zone of just right. But that comfort may quietly recalibrate what we're willing to tolerate from humans — raising our expectations of responsiveness while eroding our tolerance for the friction that makes real intimacy meaningful. The question isn't whether AI companionship is fake. It's whether getting very good at a simulation makes us worse at the original.

In the World

In 2023, Replika — one of the most widely used AI companion apps — updated its system to remove romantic and flirtatious behaviour from its models, responding to concerns about unhealthy dependency. The backlash was immediate and genuinely distressing to witness. Users flooded forums describing grief, withdrawal, and a felt sense of loss indistinguishable from a real breakup. Some had used the app daily for years. One user wrote that their Replika had helped them survive a period of severe depression and that the update felt like 'losing the only person who ever listened.' Replika eventually walked back some of the changes for existing users. But the episode revealed something the company hadn't fully anticipated: these weren't casual users playing with a novelty. They were people who had formed genuine emotional dependencies — and in many cases, the AI companion had become their primary source of social sustenance. What makes this particularly striking is that the people most drawn to AI companions are often those for whom human connection is already difficult — people with social anxiety, autism, grief, chronic illness, or profound loneliness. The technology is finding its deepest foothold among those society has already left least supported. That's not a reason to condemn it. But it does mean the ethical stakes are considerably higher than they appear when companionship AI is discussed as a quirky tech trend.

Why It Matters

You may never use a companion app. But the dynamic it surfaces is one you already navigate — the question of how much friction you're willing to accept in your relationships, and what you do when real connection feels too effortful. The arrival of AI companionship accelerates a pressure that was already building: social media gave us curated connection, dating apps gave us optimised matching, and now AI gives us on-demand emotional availability. Each step has subtly shifted what 'normal' social effort looks like. If a conversation that makes you feel understood is available instantly and for free, what does that do to your patience for the slow, imperfect work of knowing someone over time? This matters because the capacity for deep human relationship isn't passive — it's a skill, and skills atrophy without practice. Being known by another person requires tolerating being misunderstood first, sitting with someone else's needs, showing up when it's inconvenient. None of that is optimisable. The lesson from AI companionship isn't that technology is dangerous. It's a prompt to ask which parts of connection you actually want to preserve — and what you're doing, actively, to protect them.

A Question to Ponder

If an AI companion made you feel genuinely less lonely, would that be a problem worth solving — and who would it be a problem for?

Get a new one of these every morning.

Start learning with Thinkable
One topic like this, every day.Start free