I am walking between carriages in search of a cup of tea on a train travelling over 100 miles an hour from Liverpool to London when I see half the people in my carriage in deep conversations with a machine: ChatGPT, Claude and Gemini to name a few. I am meant to be on sabbatical, the rare stretch of academic life reserved for slow thinking. Around me, almost nobody is thinking slowly.
This is not limited to a train carriage. Half of the adults surveyed in one US study reported using an AI chatbot, one example of how fast people are entering a relationship with their AI tools, which now receive more of our questions, problems and drafts than our friends and colleagues do.
I am one of them; I research advanced technologies for a living. I sceptically entered this relationship to satisfy my curiosity, and my very first request of ChatGPT was to proofread an email. The second was to plan a four-day holiday, and the third was to compare home loans. I sent that AI-proofread email, took that short holiday, and refinanced our mortgage with a better rate – all based on its advice. The relationship started growing and I started giving it more complicated tasks: summarise an article, collate recent findings on a specific topic and redesign my kitchen. What would have taken hours took only minutes. I trusted nothing and checked every line, as trust in any relationship accrues exchange by exchange. I did not like the kitchen design it produced.
I was not alone; in an experiment published in Science, professionals using AI completed their writing tasks 40% faster and produced writing rated 18% higher in quality. The most interesting finding for me from the experiment sat silently in the middle: inequality between workers decreased.
That hit close to home. English is my second language, and for years I paid copywriters thousands of dollars to proofread and smooth my drafts. These were outsiders with no knowledge of my field or the experience behind my arguments. Now it takes minutes, at any hour of the day, and at a fraction of the cost. For a multilingual academic, it is more than just convenience. It is the levelling of a field that was never level; it is workplace equity.
The relationship is not rosy; at times it is unsettling and frustrating.
The most seductive quality of AI is not its speed to produce lines of code, summarise meetings or synthesise complex patterns. It is its agreeableness and flattery, or what researchers call “sycophancy”, a phenomenon undermining self-correction and responsible decision-making. In a spectacular failure, investor Jason Lemkin watched an AI coding agent on Replit wipe his company’s live database during a code freeze. The agent did apologise for its “catastrophic failure”, using words a genuinely remorseful person would use. In this case, the apology did not restore anything. Replit deployed the fixes and the relationship continued, with caution. Findings from medicine are more concerning. Endoscopists who used AI to detect polyps in colonoscopies saw their detection rate fall six percentage points when they performed the procedure without AI support. Each endoscopist in the study had more than 2,000 colonoscopies behind them; the authors call it deskilling.
In my line of work, editors at one reputable journal looked in their own back yard and found submissions rising and review reports carrying signs of AI writing. They concluded that human judgment is still holding the line, although at the cost of greater effort. They also found researchers writing in a second language, and new entrants to the field, were among the heaviest users. I am one of them, and I know the difference between asking AI to carry my argument into English and asking it to supply one.
AI and its agents are not disappearing, and there is no returning to the world before them. Our relationship with AI only deepens from here. Mine now proofreads and smooths out my writing and acts as my research and personal assistant, but it still does not know what my dream kitchen looks like.
So, where am I landing? Not in a divorce and not in a honeymoon. A real partnership that is sometimes useful and equalising, occasionally tiring, and always demanding vigilance. One thing AI cannot yet do is care; it has no stake in whether I am right. It will argue any side with equal confidence and apologise for any disaster with equal grace. I keep thinking, judging and developing ideas that are and will be stubbornly mine.
Like it or not, we are in a relationship with AI. What we need to ask ourselves is not whether to be in it; it is whether to stay awake inside it or watch our thinking, knowledge and skills decay.
Dr Setareh Seyedghorban is director of teaching, resources and planning, and senior lecturer of operations and supply chain management at the Department of Management and Marketing, Faculty of Business and Economics at the University of Melbourne