Since at least the year 1224, villagers in Törbel, Switzerland maintained communally owned property: alpine grazing meadows, forests, irrigation systems and the paths and roads that connect private and communal land. Rules included clauses such as ‘no citizen could send more cows to the alp than he could feed during the winter’.[1] Cheese was produced collectively and then doled out to those who contributed cattle. Local elected officials arranged for the distribution of manure on summer pastures. This common land in Törbel maintained its productivity for centuries.
About seven hundred years later, in 1968, an American ecologist named Garrett Hardin wrote an essay titled ‘Tragedy of the Commons’.[2] The concept had been around since antiquity, but Hardin popularised the term and it has stuck. Here is the classical image, one that looks shockingly similar — at least in structure — to Törbel and its alpine meadows: if many people have unlimited access to graze their cattle on a pasture, they will tend to overgraze and destroy the pasture altogether. Even if some people restrain themselves, others will still overgraze. Tragedy is inevitable.
We overfish, overhunt, deplete soil through overgrowing, overdraw our rivers and streams, we overpollute and we know it. But we can’t seem to restrain ourselves, or refuse to restrain ourselves if others will not. And the commons — whether it be a pasture, an estuary, a shared kitchen or a street library — does often dissolve into disrepair.
But in 1990 Elinor Ostrom published a book called Governing the Commons based on decades of fieldwork studying groups of people who managed forests and fisheries and estuaries, including the villagers in Törbel.[3] Hardin had an opinion. Ostrom had data. And Ostrom, along with many colleagues, found that sometimes people can govern the commons — even when there are quite a lot of people to organise — and can do so for many generations without depleting the resource. Governing the Commons is not as catchy as its tragic counterpart, and the details of exactly how large groups of people manage this are nuanced. But Ostrom was the first woman to win the Nobel Prize in Economics and she did it by refuting the inevitability of tragedy.
This is all preamble. What I want to talk about is poetry. And artificial intelligence. Stay with me. A few years ago, I was interning at IBM and investigating the training data for large language models; at the time, GPT-3 had just come out, followed by many other models created by other large corporations. I discovered that many of these models were trained on creative writing: bestsellers, sure, but also mid-list novels and breakout memoirs and contemporary poetry collections. There is an idea of the cultural commons: that books and music and paintings, while also existing as pieces of property, are part of a common pool resource. We all benefit from creative work. In some ways, we can do whatever we like with it, be inspired, mimic, reference or compete. Ideally, we build upon it and then contribute something of our own. Artificial intelligence, or the people and companies wielding it, draw from the cultural commons.
So do I. Recently I read Alexis Wright’s The Swan Book (2013); right afterwards, I read Ian McEwan’s What We Can Know (2025). I discussed these with my mum and my nephew in our family book club. I argued that Wright’s book was sophisticated in a way that McEwan's book was not; it attempted to capture a more difficult and nuanced idea. My mum said that Wright’s book was certainly more beautiful but was hard to comprehend compared to McEwan’s. Each book sits in a lineage, and contributes new ideas to an endless canon, and if I should ever write a novel these books will tilt my novel towards certain styles and ideas about what a novel should be.
Wright did not give me permission to read her book; neither did McEwan. One needn’t have consent from the author to read their published book. Right now, we don’t know if OpenAI or Google or Stanford needs consent to use these books to train artificial intelligence systems. Or rather, we’re debating it in the courts. Or rather, we’re debating it in the streets, and in our magazines and newspapers. I want to think of writing like an alpine meadow. I graze my cattle, but not too many, just the number I can afford to feed in the winter. I am but human, as is every novelist and poet and memoirist out there, and, as humans, we are slow. We only have so many cows. In contrast, computers have many, many cows and the companies that like to run them ragged can afford to feed them all. I don’t know if this should be illegal. But I know that we may be heading towards a tragedy, where art gets locked down for fear of scraping and artists can no longer make a living because artificial intelligence generates so much slop that real people can’t get an in.
If you’re worried about these outcomes, like I am, you might want to refuse these systems. Don’t use them, don’t pay for them, consider suing their creators or pressuring your government to regulate them. These models — whether or not their creation should have been illegal — were made by people who don't seem to care much about writers. Recently a friend shared a job listing for a ‘Writing Specialist’ posted by an artificial intelligence company. It asked for highly successful writers, defined as having ‘verified novel publishing deals with major houses ... novel sales >50,000 units ... 10+ short stories in major outlets (e.g., the New Yorker)’ to work for $40USD/hour. An insulting job post. But would a higher number have helped? Four hundred? Four thousand?
Let’s attempt to separate the people making artificial intelligence systems from the systems themselves. And let’s stop talking about artificial intelligence — a phrase whose referent changes with the wind — and instead talk about language models. Language models are a specific kind of mathematical machine that try to replicate language. Language models interest me because they wield language in an alien way, and they confuse our notion of authorship, and they spew text like even the most prolific writer can only dream of. Anyone can make a language model, although you might make one considerably smaller than those made by Meta or Bloomberg. Language models, big and small, are worth investigating, and I think writers deserve to investigate them. Can language models ethically drink from the river of our language? I have been trying to envision such a world, because refusal is powerful but also, I worry, limited. What options are we giving up by turning away?
So, I return to the commons and I remember Ostrom's work: how she refuted the many scholars who thought the commons could only end in tragedy. Writers should be governing writing-as-data themselves, not just bargaining with big corporations for better terms. Writers should be owning their work, building their own models, setting their own terms, and monitoring what their models are used for. This is not far-fetched. The technology is getting simpler, cheaper and more accessible by the day. We see new models released regularly, ones that are open-source and carefully curated.
The only people I trust to make use of the cultural commons are the cultural creators; when people manage their own resources, they also learn to protect them. Collective action is a powerful path forward. We must do this ourselves.
[1]. Elinor Ostrom, 'Analyzing long-enduring, self-organized, and self-governed CPRs’, Governing the Commons: The Evolution of Institutions for Collective Action, Cambridge University Press, Cambridge, 1990, p. 62
[2]. Garrett Hardin, 'The Tragedy of the Commons', Science, 162, No. 3859, 13 December 1968, pp. 1243-1248
[3]. Ostrom, 'Analyzing long-enduring, self organized, and self-governed CPR's', pp. 58-102