When AI Writes Our Friendships, What Do We Lose?

Midway through a funeral eulogy, a man became convinced he was listening to artificial intelligence. The telltale signs were all there: abstract nouns proliferating like weeds, formulaic assertions that the deceased was "not just X—he was Y," and a suspicious abundance of the word "collaborate" for someone describing a rec-league hockey teammate.

The eulogy was grammatically perfect yet utterly impersonal. While he had no definitive proof, the man's instinct was clear: a computer had just saved someone from the burden of thinking about their dead friend.

The Quiet Revolution in Personal Communication

Large language models are increasingly relieving people of reading and writing across all spheres of life—not just in schools and offices, but in group chats and personal emails. While institutions scramble to create AI policies and courts wrestle with intellectual property questions, our personal relationships remain the Wild West of AI adoption.

We have unspoken rules about which movies to save for our roommate and who gets invited to the lake house, but we've established nothing comparable about whether to use ChatGPT for responding to someone's Christmas letter. That oversight is becoming costly.

A November Brookings Institution survey revealed that 57% of respondents used generative AI for personal purposes, with 15-20% specifically using it for "social media or communication." What the survey didn't ask—and few statistics exist on—is whether people disclosed their AI use or passed off its outputs as their own work.

The Efficiency Trap

The people selling AI keep suggesting we use it to streamline tasks that many of us find meaningful and even enjoyable. Apple's iOS 18, for example, offers AI summaries of text message contents in notifications to make communication more "efficient."

Before turning it off, this feature summarized a group chat—where a friend shared a photo of her normally locked attic door now mysteriously ajar, sparking jokes about finally being haunted—as simply "a conversation about a wooden room." The summary wasn't just inaccurate; it stripped away everything entertaining about the exchange to reduce it to bare information transfer.

When your brother texts "How's it going?", he's not seeking data—he's seeking connection. That connection gets severed when you ask ChatGPT to draft a 50-word reply about how his baby is cute and you love him. It's equivalent to playing a recording of yourself saying "Oh, that's interesting" every time someone speaks to you.

The Economics of Care

A Microsoft survey from 2024 found that 52% of workplace AI users were reluctant to admit using it for "important tasks," presumably fearing it might make them appear replaceable. In friendly communications, that shame should be universal—but often isn't.

Generative AI creates a troubling imbalance: texts that take more effort to read than they did to write. One tutor received a 3,000-word parental response complete with section headings and bolded names—clearly generated in seconds but requiring minutes to parse. The message was clear: the parent wanted the relational benefits of a substantial reply without caring enough to write one themselves.

Writing as Proof of Work

The assumption that writing's main purpose is quick information transfer breaks down when we consider relationships defined by mutual obligations. In these contexts, written text serves as what cryptocurrency calls "proof of work"—evidence of effort expended and commitment demonstrated.

College students don't write term papers to inform professors about class dynamics in Wuthering Heights; they write to clarify their understanding for both instructor and themselves. A eulogy forces the eulogizer to think deeply about their relationship with the deceased while demonstrating ongoing commitment even when no earthly benefit remains.

Writing is an act of taking care. We shouldn't let efficiency steal that experience from us.