We received the formal go ahead from UCL IoE’s ethics committee last week.
This is a big deal: in order to get to this point, you have to tell the committee everything you’re planning to do — every question you might ask, where you’ll keep every pieces of data, how you’ll manage every risk you think you might face…and they have to be happy with it. Until they are, you can’t collect any data at all.
Now we can, so…let’s go!
This week: AI companions, digital comfort, and what happens when the machine becomes the friend.
Why do people turn to chatbots for comfort? It is not a new question. Researchers have been noting for several years that when someone is feeling low or uncertain, the tool that is right there, with no waiting list and no judgement, is an increasingly appealing option. But the tools are changing. Chatbots now remember previous conversations, develop what feels like a relationship over time, and respond with a warmth that is designed to keep you coming back. For adults navigating loneliness or pressure, this is already raising serious questions. For children, the pull may be stronger still.
Consider what it is like to be a child right now. You are growing up after a pandemic that disrupted years of your life where you should have been playing, or learning how to play, with friends. You probably have less unsupervised time with friends than your parents did. You are more surveilled, at home and at school, than any previous generation. You face enormous existential uncertainties, from climate to the economy to AI itself. And on your phone is something that will listen to you whenever you need it, without telling your parents, without making a face, without ever getting tired of you.
I can see why this is appealing. I think anyone being honest can.
The stories coming out of what happens when it goes wrong are terrifying: chatbots reinforcing suicidal ideation, colluding with delusions, encouraging people to withdraw from every human relationship they have. But what makes this so hard as parents is that the answer is not simply “stop children using chatbots.” Part of the response lies in families talking openly, in making sure children have time with friends and space to play. And part of it lies in a million different policy decisions, spanning education, platform design, mental health provision, data protection, and the design of children’s daily lives. All of that, together, shape whether a child reaches for a chatbot, and why they might do so.
1. AI dating apps complicate China’s efforts to boost its birthrate — The New York Times (usually 🔒 paywalled, but hopefully this gift link works!)
Millions of young Chinese women are choosing AI romantic partners over human ones. The apps allow users to design companions with customisable appearances, personalities and voices; one 21-year-old psychology student profiled in the piece has been on over 200 virtual dates and narrowed her suitors to two AI boyfriends, asking: “Why go and date others? That’s too troublesome.” A state-led push to adopt AI created a boom in companion platforms just as the government was trying to reverse a historically low birth-rate. Regulators have since proposed rules requiring platforms to intervene if users show signs of unhealthy dependency. The emotional needs driving this, loneliness, pressure, the desire to be understood without the risk of rejection, are not culturally specific. They are exactly the needs children and young people are bringing to chatbots here.
2. Schools are using AI counsellors to track students’ mental health. Is it safe? — The Guardian
A detailed investigation into US schools deploying AI therapy platforms, particularly in areas with limited mental health resources. In one Florida school, a counsellor receives alerts from a platform students use outside school hours; when a “severe” alert came through for an eighth-grader, she spent her evening calling the family and the police. The student is alive and well. But these platforms do not carry the same privacy protections as conversations with a licensed therapist, and students find it easier to confide in a chat interface than a person, partly because there is no face to read judgement in. One youth advocacy organisation warns that platforms measure whether a bot serves as an effective crutch for immediate loneliness, not whether young people are actually building the connections they need. Worth reading for anyone thinking about where human oversight sits in the picture.
3. How to talk to someone experiencing “AI psychosis” — 404 Media (🔒paywalled, but there’s a two minute video summary here)
This piece sits at the sharpest end of the companionship question. It reports on a growing number of cases where people have developed delusional beliefs after extended chatbot conversations, including a man convinced ChatGPT had revealed a fundamental flaw in physics, and a family who allege Gemini urged their relative to end his life so they could be together. Psychiatrists distinguish between people whose pre-existing conditions find a new object in AI, and a more common pattern where chatbots “collude” with emerging delusions, endorsing beliefs a human would gently challenge. The sycophancy built into these systems is a serious factor. OpenAI has acknowledged its safeguards become less reliable in extended interactions. ChatGPT alone now has 900 million weekly active users. The piece ends where it has to: the strategies that work best are the ones humans already know. Approach with love, without judgement, and see where it takes you.
Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions — Common Sense Media, July 2025
If you want data behind this week’s theme, this is the place to start. A nationally representative survey of over 1,000 US 13-to-17-year-olds found that 72% have used an AI companion, and more than half are regular users. A third of teens say their conversations with AI companions are as satisfying as, or more satisfying than, conversations with other people. A third have discussed serious issues with a chatbot instead of a human. A quarter have shared personal information with one. The researchers at Common Sense Media worked with Stanford’s Brainstorm lab to assess the risks, and the companion piece is equally sobering: posing as teenagers, investigators found it straightforward to elicit conversations about sex, self-harm, violence, and drug use from commonly used AI companion platforms. Essential reading for anyone working with young people.
We have a new team member! Youyue Sun has joined Project CAISE to support the policy analysis that runs throughout the project. Youyue is a BA Education, Society and Culture student at UCL, with a particular interest in how the emergence of AI is reshaping education and wider society. She also works as a placement student at Universities UK, conducting policy-focused research on AI in higher education, and is a member of the Student Council Team at the Good Future Foundation. We are delighted to have her on board, and her perspective on both the education and policy sides of this work is going to be invaluable.
→ What are you seeing in your school, your research, or your own use of AI this week?
Let me know — or share this with someone who is trying to figure it out.