AI & Kids: The Danger of Replacing Human Connection

The rapid integration of artificial intelligence into mental healthcare, offering increased accessibility and reduced clinician burnout, is prompting a crucial conversation about its limitations. Recent trials, including the first randomized controlled trial of an AI therapy chatbot, have demonstrated significant symptom reductions for depression and anxiety as reported by Infectious Disease Advisor. Yet, a growing concern centers on the potential impact of relational AI on the developing brains of young children, a risk highlighted by decades of research into attachment and social development.

The Rise of Relational AI and Youth Engagement

Data from 2025 indicates a significant increase in children’s access to and engagement with digital devices. The Common Sense Census reveals that 40 percent of children have their own tablet by age two, rising to 58 percent by age four . One in five parents already utilizes devices for emotional regulation in their children. Among teenagers, AI companion use is even more prevalent, with 72 percent having experimented with these tools and over half using them regularly. Notably, one in three teens report finding conversations with AI companions as satisfying, or even more satisfying, than interactions with real-life friends. This trend prompted concern when Mattel announced a partnership with OpenAI to integrate generative AI into children’s toys, ultimately leading to a delay in the product’s release due to advocacy group concerns.

The Cloth Monkey Experiment: A Foundation for Understanding Attachment

The core of the concern stems from research dating back to the 1950s, specifically the work of psychologist Harry Harlow with rhesus monkeys. Harlow’s experiments, while ethically complex by today’s standards, revealed a fundamental truth about attachment: comfort, not simply sustenance, is the primary driver of infant bonding. Monkeys separated from their mothers and raised with either a wire surrogate providing food or a soft cloth surrogate without food overwhelmingly preferred the cloth mother, spending the vast majority of their time clinging to it, even when the wire mother provided nourishment. This demonstrated that the need for tactile comfort and security outweighs the need for basic physical needs like feeding.

However, Harlow’s research didn’t complete there. He observed that monkeys raised solely with cloth mothers, while initially less distressed, developed significant social and emotional deficits as they matured. They exhibited reclusive behavior, social awkwardness, self-harm, and difficulties forming healthy relationships. When they became parents, many displayed neglectful or abusive tendencies. Harlow concluded that the cloth mother, while providing comfort, could not replicate the complex, responsive interaction necessary for healthy development, stating that “the nourishment and contact comfort provided by the nursing cloth covered mother in infancy does not produce a normal adolescent or adult.”

Biobehavioral Synchrony: The Missing Element

Modern neuroscience, particularly the work of Ruth Feldman, sheds light on why a cloth mother – or, by extension, an AI companion – falls short. Feldman’s research highlights the importance of “biobehavioral synchrony” – the synchronized alignment of heart rhythms, oxytocin release, and brain oscillations between individuals during face-to-face interaction. This synchrony is not merely a byproduct of connection; it’s the mechanism by which a mature brain helps regulate and shape the developing brain of an infant, tuning it for social interaction. This synchrony occurs across various attachment types – parent-child, romantic partners, even close friends – and is demonstrably reduced when interaction is mediated by technology as Healio reports.

Humans are fundamentally social creatures, wired for connection through biological signals that algorithms cannot replicate. The quality of that connection, the responsiveness and attunement, is critical for healthy brain development.

A Simple Test for AI in Relationships

Despite the potential benefits of AI in healthcare – increasing accessibility, reducing clinician burnout, and even potentially fostering real-world connections – a critical question must be asked of any AI designed to operate in the realm of human relationships: Does it lead people toward other people, or away from them? For young children, in particular, tools that mimic relationships, offer open-ended conversation, or present themselves as human should be approached with extreme caution. Developmentally beneficial tools should instead encourage real-life interactions and positive actions that promote thriving.

The discomfort many feel about children forming emotional bonds with machines is not simply alarmist; it’s rooted in decades of scientific research. AI that strengthens human bonds can be a valuable tool. AI that substitutes for them, especially for developing children, risks repeating the lessons learned from Harlow’s experiments – creating a “cloth monkey” at scale.

We’ve already conducted this experiment, and the results are clear. We don’t need to repeat it.

Looking Ahead: Prioritizing Human Connection

The conversation surrounding AI and mental health, particularly for children, needs to shift from what’s technologically possible to what’s developmentally appropriate. Ongoing research, like Feldman’s work on biobehavioral synchrony, will continue to refine our understanding of the biological underpinnings of social connection. Regulatory bodies and tech companies must prioritize ethical considerations and long-term developmental outcomes over short-term gains. The goal should be to leverage AI to enhance, not replace, the vital human connections that are essential for healthy growth and well-being.