Who Keeps the Golem?

<h2 id="h-the-aspen-institute-has-released-a-new-white-paper-about-ai-authorship-and-what-is-at-stake-for-the-future-of-books" class="wp-block-heading has-h5-variant"><em>The Aspen Institute has released a new white paper about AI, authorship, and what is at stake for the future of books.</em></h2> <p class="wp-block-paragraph"></p> <p class="wp-block-paragraph">There is an old story about a rabbi who makes a man from river clay and brings him to life with a written word. The golem works without rest and follows orders. It carries water and protects the community. But it cannot pray, have meaning, or take responsibility for its actions. When it becomes dangerous, the rabbi erases one letter from the word <em>emet</em>, which means truth, turning it into met, <em>meaning</em> death. The golem then returns to mud.</p> <p class="wp-block-paragraph">Now we have created a new kind of golem. Instead of carrying water, it writes. This machine has been trained on much of what people have published, and it can produce memos, poems, and even summaries of books it has never truly read. It can even mimic the writing of those who have passed away. Its existence has reopened an old question in literary theory: what does it mean to call someone an author? In my white paper for the Aspen Institute, “Beyond the Algorithmic Author,” I explore this in three parts.</p> <figure class="wp-block-image alignleft size-medium is-resized" id="Beyond-the-Algorithmic-Author"><a href="https://www.aspeninstitute.org/wp-content/uploads/2026/07/Beyond_the_Algorithmic_Author_R-Santodomingo-The-Aspen-Institute.pdf"><img loading="lazy" decoding="async" width="621" height="800" src="https://www.aspeninstitute.org/wp-content/uploads/2026/07/Beyond-the-Algorithmic-Author-621x800.png" alt="Essay's cover: Beyond the Alorithich Author" class="wp-image-263108" style="aspect-ratio:0.7762132646174612;width:294px;height:auto;object-position:50% 50%" title="Link to the full document donwload"/></a><figcaption class="wp-element-caption">Download the White Paper <a href="https://www.aspeninstitute.org/wp-content/uploads/2026/07/Beyond_the_Algorithmic_Author_R-Santodomingo-The-Aspen-Institute.pdf">Beyond the Algorithmic Author</a></figcaption></figure> <p class="wp-block-paragraph">The first part looks at the purpose of authorship. The paper argues that an author must be a person who can mean what they write and take responsibility for it. Machines, no matter how well they imitate people, cannot do this. The law agrees for now: in March, the Supreme Court chose not to hear <a href="https://www.scotusblog.com/cases/thaler-v-perlmutter/"><em>Thaler v. Perlmutter</em></a>, so the rule remains that only humans can be copyright authors. The main reason is accountability. A machine cannot go to court, intend to harm someone’s reputation, or feel regret. The paper suggests moving past the old question of whether a machine can fool us, and instead asks a new one: could the author regret what they wrote? Regret requires a real self with something to lose. The golem cannot pass this test, and this is what readers really want to know when they ask if a human wrote something.</p> <p class="wp-block-paragraph">The second follows the money. Last September, the value of a book to a machine became clear when Anthropic agreed to pay $1.5 billion, <a href="https://www.publishersweekly.com/pw/by-topic/digital/copyright/article/100438-little-drama-at-anthropic-s-settlement-hearing.html">the largest copyright settlement in U.S. history</a>, to authors whose books were taken from pirate libraries. That comes out to about three thousand dollars per book, and it is important to be clear about what that number means. A court had already decided that using legally bought books for training was fair use; the settlement was about the stolen books. By putting a price on this, it turned the shared world of books into something that can be bought and sold, which creates a problem the paper explores in detail. The more we focus on consent and payment for training data (which is important), the more we give control of “ethical” AI to a few huge companies that can afford it. Solving this problem will require more than just lawsuits.</p> <p class="wp-block-paragraph">The third belongs to the reader, and this spring supplied the evidence. In May, a book about how AI corrodes our grip on truth turned out to be studded with quotations that the author’s AI tools had made up. The book was polished enough to get endorsements and pass fact-checkers, but that polish is now a problem. Readers have always trusted polish as a sign of care, but it no longer means what it used to. Soon after, the Caribbean winner of the Commonwealth Short Story Prize was accused online of using AI to write his story. The organizers did not use detection software, which they found unreliable. Instead, they carefully checked the writer’s drafts, timestamps, and notes, and cleared him. In July, he was <a href="https://www.thedailystar.net/culture/books-literature/news/news/jamir-nazir-wins-2026-commonwealth-short-story-prize-following-ai-review-4214976">named the overall winner</a>. He proved his work was his own by letting others see his writing process, but the prize’s longtime publisher still ended their partnership. It is easy to make an accusation, but being cleared costs the writer his privacy and costs the prize its publisher.</p> <p class="wp-block-paragraph">I want to share my own position on these issues, just as I do in the paper.</p> <p class="wp-block-paragraph">English is my second language. I also drag words through the rough trails of <a href="https://www.cicero.forum/p/me-dyslexic">dyslexia</a>. Every page I write in English is a kind of translation, and I use a program called Grammarly to help with copyediting, though the ideas and decisions are my own. This means I understand the need for these tools, but I also know the suspicion they create. Writers like me, who are careful in a language we learned later, are often the first to be flagged by detection tools. That is why the paper does not trust forensic methods and instead relies on openness. If a writer signs off on machine-written text without saying so, they become what Spanish calls a <em>prestanombres</em>, a <em>testaferro</em> of words, reducing their craft to a name used for someone else’s work. Being honest about how a text was made keeps the writer’s name connected to their own work.</p> <p class="wp-block-paragraph">Big problems often lead to big solutions, but the paper ends with a simple one. It suggests creating a permanent working group on AI and books that convenes large publishers, small presses, author and translator groups, labs, librarians, and lawmakers. Success would be measured by practical tools, not just statements: model contract language for training rights, clear disclosure standards for libraries, a plan for collective licensing that is affordable for more than just a few, and a short yearly report on where private actions have gone beyond the rules. The institute already has much of what is needed. Aspen Digital has started these conversations, and the new Technology Leaders Initiative will bring together senior technologists focused on learning, not just promoting their own views. Those who work with literature in translation should also be included to ensure that voices beyond English are not forgotten.</p> <p class="wp-block-paragraph">I close the paper with this thought: a machine can now produce more words than any writer in history, but it has less voice than any of them. It has read everything but experienced nothing. A machine cannot tell us what it feels like to be human, just as a photograph cannot show a painter’s inner world. That is the writer’s job, and the reason readers look for real writers. The golem will keep writing, as it can. But deciding who is responsible for it, who agrees to it, and who signs their name to it is not something a machine can decide. That choice belongs to us.</p> <p class="wp-block-paragraph"><em><a href="https://www.aspeninstitute.org/wp-content/uploads/2026/07/Beyond_the_Algorithmic_Author_R-Santodomingo-The-Aspen-Institute.pdf">Download the full white paper</a></em>.</p> <div class="wp-block-media-text is-stacked-on-mobile"><figure class="wp-block-media-text__media"><a href="https://www.aspeninstitute.org/blog-posts/the-artist-is-still-in-the-room/"><img loading="lazy" decoding="async" width="1500" height="831" style="object-position: 50% 50%" src="https://www.aspeninstitute.org/wp-content/uploads/2026/07/AI-and-creativity-1500x831.png" alt="" class="wp-image-263105 size-full"/></a></figure><div class="wp-block-media-text__content"> <p class="has-medium-variant wp-block-paragraph"><a href="https://www.aspeninstitute.org/blog-posts/the-artist-is-still-in-the-room/">The Artist Is Still in the Room</a></p> <p class="wp-block-paragraph"><em>The case for practical optimism. A dispatch from Shared Futures: the AI Forum, at the Times Center.</em></p> </div></div> <div class="wp-block-post-author"><div class="wp-block-post-author__avatar"><img alt='' src='https://secure.gravatar.com/avatar/ad4c52f6e6a63abf097ecd3d85b9d3e4bcfe06fa7c334880b388fcbe752c7171?s=48&#038;d=mm&#038;r=g' srcset='https://secure.gravatar.com/avatar/ad4c52f6e6a63abf097ecd3d85b9d3e4bcfe06fa7c334880b388fcbe752c7171?s=96&#038;d=mm&#038;r=g 2x' class='avatar avatar-48 photo' height='48' width='48' /></div><div class="wp-block-post-author__content"><p class="wp-block-post-author__byline">The Aspen Institute</p><p class="wp-block-post-author__name">Roger Santodomingo</p></div></div> <details class="wp-block-details is-layout-flow wp-block-details-is-layout-flow"><summary>AI use</summary> <p class="wp-block-paragraph">This article has been copyedited using Grammarly, which is an AI language tool.</p> </details> <p class="wp-block-paragraph"></p> <p>The post <a href="https://www.aspeninstitute.org/publications/who-keeps-the-golem/">Who Keeps the Golem?</a> appeared first on <a href="https://www.aspeninstitute.org">Aspen Institute</a>.</p>