On a Train at 100mph, an Academic Sees Everyone Talking to AI

Half the passengers in Setareh Seyedghorban's train carriage were deep in conversation with a machine. Writing in The Guardian, the multilingual academic describes walking between carriages in search of a cup of tea on a train travelling over 100 miles an hour from Liverpool to London — and finding people engrossed in conversations with ChatGPT, Claude and Gemini.

She was on sabbatical. That is, in her words, "the rare stretch of academic life reserved for slow thinking." Around her, almost nobody was thinking slowly.

That single scene anchors her essay, titled "My (complicated) relationship with AI: often seductive, occasionally frustrating and always demanding vigilance." It is a first-person account of how a working academic now lives alongside chatbots — and what that coexistence costs and returns.

The scale behind the anecdote

Seyedghorban's observation is not confined to one carriage. She cites a US survey, reported by NBC News, in which half of American adults said they had used an AI chatbot. That figure, she argues, illustrates how quickly people are entering into a relationship with their AI tools.

The stakes, as she frames them, are relational as much as technological. These systems "now receive more of our questions, problems and drafts than our friends and colleagues do." A tool that began as a curiosity has become, for many users, a first stop for thinking itself — before any human interlocutor is consulted.

A levelled playing field

Seyedghorban's relationship with AI is not hostile. As a multilingual academic, she writes that artificial intelligence "has levelled the playing field." For scholars whose working language is not their first, chatbots offer a form of practical equity: help with drafting, refining and expressing ideas that native speakers once accessed through editors, colleagues or privilege.

But the benefits arrive bundled with demands. Her subtitle characterises the relationship in three registers: "often seductive, occasionally frustrating and always demanding vigilance." Each word carries weight. The seduction is the ease. The frustration is the failures. The vigilance is the permanent tax she believes the tools impose on their users — the obligation to check, to verify, to push back against plausible output.

Her conclusion is a declaration of intent rather than a verdict on the technology. She writes that she will "continue to think, judge and develop ideas that are stubbornly mine." The adverb does the work: the ideas are not merely hers, but stubbornly so — held against a current that makes outsourcing thought frictionless.

Why this column matters now

First-person accounts from academics have become a distinct genre in the AI debate, and they carry information that benchmark scores do not. Adoption statistics measure usage; essays like this one measure texture. Seyedghorban describes not whether people use ChatGPT, Claude or Gemini, but how it feels to use them daily while trying to preserve the habits of a scholar — drafting, doubting, revising, and arriving at conclusions through effort rather than prompt.

The context sharpens the column's edge. She composed these observations during sabbatical, the one period academic life explicitly protects for unhurried thought, on a train where the machine-mediated conversation had already crowded out the window stare. The survey figure she cites — half of US adults — suggests her carriage was not an outlier but a preview.

Her framing also sidesteps the two loudest positions in the public debate. She neither celebrates AI as liberation nor condemns it as decay. She treats it the way one treats a powerful collaborator with a known reliability problem: useful, uneven, and never to be left unsupervised. "Always demanding vigilance" is the operative phrase — a maintenance cost she accepts because the returns, for a multilingual researcher, are real.

The so-what

The column ends on her commitment to thinking that remains "stubbornly" her own, which is less a resolution than an ongoing practice. As half of surveyed US adults settle into routines where chatbots receive their questions first, Seyedghorban's account offers the terms of a workable settlement: use the tools, audit the output, and refuse to let the drafting ease become the thinking itself. Whether that settlement holds as the tools improve — and the vigilance they demand grows harder to sustain — is the question her column leaves open.