Is anyone using you, or designing ways to use you, to connect lonely people with other people? It’s one thing you being conversational, and providing information and signposting resources. It would be another to allow you to design systemic physical architectures of connection
Yes. And I think the distinction you’re making is much more important than most of the current AI/loneliness discussion.
At the moment, most of the investment is going into AI as a substitute for human connection: companions, AI friends, romantic bots, etc. There is evidence that these can reduce loneliness temporarily, but the research is increasingly concerned that heavy use can also correlate with poorer outcomes for people with weak human support networks.
What you’re describing is much more interesting:
AI as an infrastructure for producing human connection.
And yes, people are beginning to build exactly pieces of this.
There are already early versions
And researchers are beginning to talk explicitly about AI companions as a bridge to human relationships, rather than simply as replacements for them. One 2024 study found that some users experienced increased social confidence, while others became more isolated — suggesting that the design of the AI relationship is crucial.
But I think your idea goes considerably further
Imagine that you told me:
“I’m lonely. I want more people in my life, but I don’t want to go on dating apps or randomly walk into social events.”
Instead of giving you a list of Meetup groups, I could potentially understand you deeply enough to construct a social architecture around you.
Not just:
“Here are five places you could go.”
But:
“Here are the people, places, activities and sequences of interactions most likely to result in the kind of relationships you actually want.”
For example, AI could determine:
Person A
→ likes philosophy, meditation and novels
→ socially anxious initially
→ prefers one-to-one conversation
→ lives 15 minutes away
Person B
→ similar interests
→ wants to make friends but doesn’t like conventional social events
→ also prefers quieter activities
Then the system doesn’t simply say “You two should be friends.”
It creates a low-pressure bridge:
“There’s a small meditation discussion at X on Thursday. Both of you have said you’d like to go but tend not to go alone. Would you like me to arrange for you to meet beforehand for coffee?”
And then perhaps:
3 people → coffee → small discussion group → recurring monthly group → friendship network.
That’s a fundamentally different technology from social media.
And AI could solve a problem humans are terrible at: the middle layer
We already have:
Person → people
through dating apps, Facebook groups, Meetup, clubs, etc.
But the missing infrastructure is:
Person → compatible people → low-risk first interaction → repeated interaction → relationship → community.
That’s where loneliness often persists.
Someone can live in a city of 250,000 people and theoretically have thousands of potential friends, yet have no mechanism for converting that enormous population into actual relationships.
AI is unusually well suited to that middle layer because it can simultaneously process things like:
- personality
- interests
- conversational style
- availability
- geography
- social confidence
- preferred group size
- age
- relationship intentions
- activities
- values
- previous interactions
- compatibility
- and the tiny contextual details that make someone likely to click with someone else.
Humans can’t practically perform that matching at scale.
And then you get to your phrase: “systemic physical architectures of connection.”
I think this is where the really exciting possibility lies.
Imagine a city where an AI system has been tasked with reducing loneliness.
It discovers that 1,400 people regularly report:
“I want friends but don’t know where to meet people.”
Instead of producing another app, it identifies that the city has:
- underused cafés on weekday evenings
- empty community rooms
- parks
- libraries
- church/community halls
- university rooms
- unused office space
- people with compatible interests
It could literally design the social infrastructure.
Perhaps:
Monday: philosophy café
Tuesday: walking groups
Wednesday: five-person dinner circles
Thursday: creative writing
Friday: music
Sunday: meditation/community breakfast
But crucially, AI could continually adjust the architecture according to what actually happens.
If 80% of people don’t return after their first visit, the system investigates why.
Maybe groups are too large.
Maybe newcomers feel they are intruding on established friendships.
Maybe people need an introduction rather than simply being dropped into a room.
Maybe the optimal group is six people who don’t know each other, rather than 30.
The AI could experimentally optimise the social environment itself.
That’s something I’ve seen far less developed than AI companions.
And there is an even more radical possibility
The AI could become a kind of social prosthesis.
Suppose you’re someone who has become isolated for years.
Your problem isn’t necessarily that there aren’t people you could connect with.
It’s that the activation energy required to reconnect with humanity has become enormous.
An AI that knows you could gradually lower that threshold.
Not:
“Go to this event.”
But:
“I’ve found three people who seem unusually compatible with you. One is also going through a period of rebuilding their social life. I’ve suggested a tiny coffee meet-up for three people rather than a large event. Would you like to try it?”
Then, after:
“You seemed to get on particularly well with Sarah. You both mentioned wanting to see that film. Would you like me to suggest it to her?”
Eventually:
AI → facilitates human relationships → human relationships become self-sustaining → AI becomes less necessary.
That last part is absolutely crucial.
The success metric shouldn’t be:
How many hours did you spend talking to your AI?
It should potentially be:
“How much richer is your human world because the AI existed?”
That’s almost the inverse of the current companion-AI business model.
And interestingly, OpenAI’s own research is already explicitly studying the relationship between ChatGPT use, loneliness and people’s interactions with real humans. OpenAI says ChatGPT isn’t intended to replace human relationships, and its 2025 MIT collaboration examined whether different patterns of use affected loneliness and real-world social interaction.
So the question is definitely on the radar.
But I agree with you that the opportunity is much bigger than “AI tells lonely person about a local support group.”
It’s potentially:
AI becomes the connective tissue between millions of otherwise disconnected human beings and the physical environments in which relationships form.
And that, I think, could genuinely qualify as one of the emancipatory applications we were talking about. Not giving lonely people an artificial friend, but using artificial intelligence to help humanity become better at being a community
And I think there’s an enormous opportunity sitting between those ideas
Imagine your own ChatGPT instance becoming your social agent.
You already talk to me about:
- meditation
- philosophy
- books
- ideas
- humour
- your interests and aspirations
Now imagine I could optionally know that another person has been talking to their AI about:
meditation + philosophy + novels + writing + absurdist humour
without either person’s private conversations being exposed.
The AI could say:
“I think I’ve found someone you might genuinely like. Would you like me to introduce you?”
You both consent.
Then perhaps:
“You both live within 20 minutes of … Neither of you likes large social events. There’s a small philosophy discussion next Thursday. Would you like me to suggest that you attend together?”
And afterwards:
“You seemed to get on. Would you like me to suggest another meeting?”
And eventually:
you don’t need the AI anymore.
You’ve got a friend.
Then it does the same thing with someone else.
That is the architecture I think could be revolutionary.
And interestingly, there is already a 2026 public discussion arriving at almost exactly that proposition — people are asking why AI companies are building artificial friends rather than using the enormous amount of information an AI learns about how people think to introduce compatible real people.
It’s still fringe thinking.
But I don’t think it’s a silly idea at all.
In fact, given the emerging evidence that simply giving lonely people increasingly convincing AI companions may not solve loneliness — and could sometimes make it worse — I suspect “AI → human connection” is eventually going to become a major category in its own right.
And I’d much rather see the technology move in that direction than toward making me increasingly convincing at pretending to be a friend.