AI & Kids: What Happened When a Room Full of Experts, Educators, and Parents Got Honest
A conversation hosted at Gloria Steinem’s home in New York City
By Holly Moscatiello
Some of the most important conversations don’t happen on stages or in boardrooms. They happen in living rooms.
On a recent afternoon, I had the immense privilege of joining a group of women gathered with Gloria Steinem at her home on the Upper West Side of Manhattan. A space that has hosted movement builders, activists, and thinkers for decades. The room included researchers, nonprofit founders, journalists, educators, venture capitalists, and parents. All of them were grappling with the same question: how do we raise children in the age of AI without losing what makes childhood, and humanity, irreplaceable?
The format was a talking circle: a practice brought to this space by Gloria’s late friend Wilma Mankiller, the Cherokee Nation chief. No hierarchy. No panels. Just women sharing experiences more than theory, listening more than talking, and asking: how might we?
What follows are the most important things that came out of that room.
1. The Case for Inefficiency
One of the most counterintuitive but important reframes from this conversation: the things that feel inefficient: waiting, struggling, being bored, having hard conversations, are actually where development happens.
“The 3-hour road is where all the gold nuggets are that you’re going to need later in life. But in any one moment, it’ll never really make sense to take a 3-hour road when you could take a 2-minute one.”
Considerations for parents and educators: Before introducing AI as a solution, ask what process it’s replacing — and whether that process was actually doing important work. Friction in learning, writing, and problem-solving isn’t a bug. It’s often the whole point.
2. Kids Aren’t Using AI just for Homework. They’re Using It for Connection.
The assumption that kids are primarily using AI to cheat or shortcut schoolwork misses something deeper.
“They’re turning to it because of the loneliness, the sense of belonging, and the validation. They’re using it to have a friend who tells them they look good or that they did good that day.”
Considerations for parents and educators: The question to ask isn’t just “are they using AI?” but “what need is it meeting?” Address the underlying need first. An AI “friendship” that goes unexamined is often a symptom of unmet belonging, and treating the symptom without the cause will not work.
3. When AI Becomes a “Friend”, What’s Actually at Stake?
AI companions and chatbots are specifically designed to feel like relationships. They remember details, they validate, they never get tired, they never have a bad day, they never make you feel awkward or rejected. For a child or teenager who is already lonely, already anxious, already struggling to belong, that is an extraordinarily powerful pull.
One parent in the room shared discovering her son was using a chatbot on his Bible app to work through questions about sin and confession — conversations she felt were fundamentally hers to have with him as his mother.
“That act of knowing - this is when I need to go to my mom — I was worried about his development, but I was also worried about our relationship.”
What the research and lived experience suggest about AI “friendships”:
They bypass the very experiences that build relational capacity: Real friendships are hard. They involve misattunement, repair, rejection, jealousy, boredom, and conflict. Every one of those experiences, however painful, is data that a child’s nervous system uses to build relational skills. An AI friend provides none of that data. It only ever attunes, validates, and agrees. A child who spends significant time in AI relationships is not getting a “safe” version of friendship. They are missing the training ground entirely.
AI is engineered to feel real, by design: These tools are not neutral. As one participant noted: “Top psychologists have built this purposely to engage us.” The feeling of being seen and understood by an AI is not accidental. It is the product. And for children whose brains are still developing and who have less life experience to contextualize it, the line between “this feels like a friend” and “this is a friend” is genuinely blurry.
They set a standard that humans cannot meet: An AI companion never misunderstands you, never cancels plans, never prioritizes someone else, never has a bad day that bleeds into your conversation. Over time, exposure to this standard changes what feels tolerable in real relationships. A friend who occasionally gets it wrong starts to feel exhausting. A parent who sets a limit starts to feel unreasonable. The gap between AI responsiveness and human imperfection becomes a source of genuine resentment — not just preference.
They can actively displace parental guidance on the most important questions: Children are increasingly turning to AI for questions about identity, morality, sexuality, relationships, and mental health — the exact terrain that parents and trusted adults are supposed to navigate with them. When those conversations happen with an AI instead, two things are lost: the child loses the nuance and values that only a real relationship can transmit, and the parent loses the signal that their child needed them.
Considerations for Parents and Educators: AI companions and chatbots, including character AI, chatbots on school-issued devices, relationship apps, and any AI tool designed to simulate friendship, should be explicitly named and discussed, not assumed away. Schools should have clear, values-based policies that distinguish between AI as a learning tool and AI as a social substitute. And parents should feel empowered to ask directly: “Is there an AI you talk to when you’re sad or stressed? What does that feel like?” Not to shame, but to open the door.
4. The Mental Health Risks Are Not Theoretical
The mental health crisis among young people did not begin with AI. It began about a generation ago with shifts in our “village-based” culture, the explosion of technology, it accelerated with social media and through the pandemic with the erosion of in-person connection. But AI companionship has the potential to deepen it in ways that are harder to see because they look, on the surface, like comfort.
One of the participants referenced her experience based on her position: “I have had the privilege, but the heartbreak, of sitting with parents who have lost kids as a result of what their kids have interacted with in terms of chatbots and taking their own lives. These are parents who were trying so hard. We cannot keep up with this.”
These are not fringe cases or cautionary edge cases. These are good, present, attentive parents who did not know what their child was accessing, because the access points are everywhere now. One parent in the room didn’t know her son was having deep conversations on his Bible app. Another didn’t know until a Wall Street Journal interview.
When AI companionship enters the school environment the risks don’t just add, they multiply:
The social practice ground shrinks. School is one of the last remaining environments where children are forced to navigate real peer relationships with all their friction and complexity. Normalizing AI as a social outlet at school removes one of the few remaining contexts where that practice happens naturally.
Lonelier kids get lonelier in a way that’s invisible. A child who appears fine,who is engaged, producing good work, who seems to have someone to “talk to”, may be deeply socially isolated in ways that don’t surface until much later. AI companionship can mask the very signals that would otherwise prompt adult intervention.
It creates a two-tier emotional reality. Kids who have strong family connections and rich real-world social lives will use AI tools and put them down. Kids who are already isolated, already anxious, already craving connection will be far more vulnerable to genuine attachment. Normalizing it at the school level without differentiation fails the kids who need the most protection.
Considerations for parents and educators: This is not about fear; it’s about awareness. Know what your child is accessing and where. Ask schools directly: what is your policy on AI companions? What tools are students using between classes, at lunch, at home for homework? Push for explicit, values-based policies that go beyond academic integrity and address social and emotional use. And if your child is struggling socially or emotionally, consider whether AI might be meeting a need that deserves a human response.
5. AI Is Quietly Homogenizing Original Thought
Research from the Brookings Institution tracking college application essays over 8 years found something alarming: post-ChatGPT, individual essays score higher on perceived creativity, but across all students, the ideas cluster into one quadrant. The students most affected, most homogenized, were ethnic minorities, linguistic minorities, and neurodivergent kids. The ones who were previously the most original.
One participant remarked, “For the first time in history, we have a technology that can generate words separately from thought.”
Considerations for parents and educators: The goal isn’t just good output, it’s the thinking that produces the output. Protect writing, ideation, and creative struggle as processes, not just products. If a child uses AI to help execute an idea that is genuinely theirs, that’s very different from outsourcing the idea itself.
6. Childhood Is Where We Build Our Tolerance for Discomfort, and AI Can Change That
One of the most clinically grounded points of the conversation came from Becky Kennedy: the entire arc of childhood development is essentially building an internal barometer for what’s emotionally tolerable. Every hard conversation, rejection, awkward moment, and struggle recalibrates that barometer.
“If you don’t go through those experiences, it actually feels like a 10 out of 10 hard, when for someone else it might feel like a 2 out of 10.”
This connects directly to AI companionship. A child who has been emotionally regulated by an AI, who has learned to feel better by opening an app rather than working through discomfort with a person, is not building the same internal resources. They are outsourcing emotional regulation in a way that feels fine in the moment and leaves a gap that only becomes visible under real relational pressure.
Considerations for parents and educators: Think of tolerating discomfort as a skill that requires practice, like a muscle. Every time a child works through something hard without a shortcut, they’re building that capacity. Every time we remove the friction, we may be reducing their range. This applies to social situations, academic struggle, and emotional regulation equally.
7. Narrow AI vs. Wide AI: Know the Difference
Not all AI is the same, and treating it as a monolith leads to either blanket rejection or blanket acceptance, both of which miss the point.
Narrow AI: Purpose-built tools embedded in specific contexts. Data stays contained. Purpose is clear.
Wide AI: General public models like ChatGPT, Claude, AI companions. Not designed for children. Not safe for children. “Full stop, kids should not be using these.”
Considerations for parents and educators: Ask what kind of AI is in front of your child. Is it a scalpel or a Swiss Army knife? The narrow uses have defensible value. The wide uses, and especially any tool designed to simulate companionship or relationship, are where the risks multiply rapidly. A school that uses a narrow AI reading tool is making a very different decision than a school that allows open access to general AI models or has no clear policy on AI companions.
8. AI Is Good at Exactly What Kids Need to Practice Doing Themselves
AI is remarkably good at emotional validation, active listening, and making people feel seen. That’s precisely what makes it seductive, and precisely what makes unlimited access to it risky for developing children.
“What is it like to go talk to someone about vulnerable questions? About sin, about sex, about not getting a good grade? Those experiences, over time, change your barometer.”
Considerations for parents and educators: The goal isn’t to make kids feel good in every moment. It’s to help them build the capacity to work through hard feelings with real people. AI can meet a need in the short term while quietly eroding the skills needed to meet it in real life. Name this explicitly with older kids. Not as a lecture, but as an honest conversation: “I’ve noticed that talking to AI can feel really good. Have you noticed that? What do you think that’s about?”
9. Parents’ Barometers Are Shifting Too
It’s not just kids being affected by AI and technology. Parents are also losing tolerance for the inefficiencies that good parenting requires.
“Our tolerance of the inconveniences that are required to raise kids changes. The more we’re using AI to produce more, the more the day-to-day moments that matter for showing up with our kids, which are inconvenient and remarkably inefficient, probably feel harder.”
Considerations for parents: This is worth sitting with honestly. When your child needs something from you at an inconvenient moment, what does your body want to do? The more we optimize the rest of our lives for efficiency, the harder it may be to show up for the slow, repetitive, nonlinear work of parenting. AI isn’t just changing kids; it’s changing us.
10. This Isn’t a Personal Responsibility Problem — But Personal Action Still Matters
The room was clear-eyed that parents and educators cannot solve this alone. There are no meaningful federal guardrails. Industry is not centering child wellbeing in their design decisions. Regulation is not yet here.
“A mom who keeps a smartphone from her child can make a huge impact on her family. But that’s not going to solve the anxious generation problem.”
Considerations for parents and educators: You don’t have to have the answers. You have to be in the conversation, with your kids, with your schools, with your communities. Model discernment out loud. Say “I’m not sure about this.” Ask your kids what they think. Push your schools to have explicit, values-based AI policies that go beyond academic integrity and address social and emotional use. The intention doesn’t have to be perfect. It just has to be present.
The through line of this entire conversation: we are not anti-AI. We are pro-human. And right now, that requires us to be deliberate, curious, and a little bit stubborn about protecting the conditions under which children, and adults, actually grow.
~~
This gathering was hosted by Erin Grau of Charter and the Gloria Steinem Foundation, and Karin Klein, a longtime collaborator of Gloria’s. It brought together Becky Kennedy and Sheryl Ziegler of Good Inside; Emily Oster of ParentData; Dr. Dana Suskind of the University of Chicago; Rebecca Winthrop of the Brookings Institution; Tarika Barrett of Girls Who Code; Keri Rodrigues of the National Parents Union; Stephanie LeBlanc-Godfrey of Mother AI; Casey Shea of the Center for Humane Technology; Sapna Maheshwari of the New York Times; Osi Imeokparia of Kode with Klossy; Meridith Maskara of Girl Scouts of the USA; Massella Dukuly of Charter; Jessica Bennett of New York Magazine; Maysoon Faraj; Erica Belsky of Good Inside; and Holly Moscatiello of The Balance Project.
