Project Panama: books are food for the...
Thoughts | Amanda He
Graphic by Cassie Hui
Beginning in early 2024, Anthropic, one of the world’s leading AI firms, began a covert effort to teach its Claude “how to write better.” The initiative targeted books predating the corruption of generative AI, purchasing tens of thousands of them from multiple vendors and outsourcing the spine-cutting and page-slicing so the content can be processed in bulk using industrial imaging equipment. Eventually, Anthropic created a digital ether with which to train large language models (LLMs). Project Panama is what they called it: a secret plan “to destructively scan all the books in the world.”
Sounds pretty bad, doesn’t it?
One could imagine that people’s first reaction to such a project is likely one of shocked disgust. Books have traditionally been treated as objects worthy of preservation, and there is a sense of sacrilege attributed to damaging them. However, the issue—or the lack thereof—of Project Panama is rather straightforward in court.
A group of book authors initiated a class action regarding copyright against Anthropic PBC in August 2024. They claimed that making copies of their books to train LLM is an act of infringement, while Anthropic argued it to be fair use. The Northern District Court of California was clear in their ruling:
District court holds that Anthropic’s use of books to train its Claude large language models and its use of purchased copies of books to create digital permanent library constitute fair use, but its use of pirated books to create such library does not constitute fair use.
—Bartz v. Anthropic PBC
Indeed, the purchases of the print copies were legally “fair and square,” and each purchase gave Anthropic the right to dispose of the copy as it saw fit. The destruction of books can perhaps be described to be as enraging and as not morally wrong as using a portrait your friend painted you as a doormat. The actual unlawful action by Anthropic was its use of pirated books from online sources in its digital library, which then became a rather simple problem of infringement on intellectual property.
Anthropic eventually agreed to pay $1.5 billion to settle the case, covering about 500,000 eligible works. Per the authors’ lead attorney Justin Nelson, it is the largest known copyright recovery in history. At the same time, it is also one of the first cases in this wave against AI companies to settle.
Is it a win, or a loss? Difficult to say.
The more interesting question is: What does it mean now that we are willing to destroy literature to feed machines, both of which are our finest creations? I believe the answer is not so callous and reductionist as a simple hunger for profit, like many seem to suggest.
Perhaps, then, Project Panama presents us with a disturbance beyond copyright. Anthropic did not destroy books because they didn’t matter, but rather they mattered enough to be useful in the creation of something potentially revolutionary; what is written by a human becomes nourishment for machines created by another human.
Have we finally reached the cusp of another stage of humanity where we have truly begun to perceive the “creation of a mind” as a reality? Compared to AI, we have so far retained the upper hand in literature, art, emotionality—things that Dostoevsky would describe as “beautiful and lofty,” things that make us human. Do we now have faith in the ability of machines to not only surpass us in computational power—which it did long ago—but to do so in the vast field of liberal arts as well in the near future? Will that ever be possible?
Humans have created art since the very beginning. Passed down oral stories, developed writing systems, saw the world through the eyes of Venus de Milo and the Great Sphinx of Giza and the Terracotta Army entombed in mud. But art has always been about the same shared experience, the same emotions, the same pleasure and the same pain, transient across time; it is as much a beauty as it is a limitation. Art evolves with us, rises and falls with society, but it is not within its nature to erupt. It is difficult to imagine a day where we discover a new emotion, something so heartwrenching that it somehow elevates our understanding of life forever. The evolution of technology, however, is an eruption. Machines lay dormant for millenia, waiting for science to branch further and further from religion and philosophy until the day they emerge from our minds to turn the world upside down. Ideally, there is a check-and-balance between the liberal arts and technology, where the majority of the world is still literate, sophisticated, and wise enough to handle something so unique like artificial intelligence. Unfortunately, the opportunity to produce that balance has long passed. We are already seeing it: the stagnation of the human mind versus the ever-rapid growth of the mechanical one. No proposal of slowing down AI development will work now; mankind has discovered fire and will thus play with it. After all, morbid curiosity has always been the defining trait of humanity.
Though some of those books Anthropic destroyed were rare, it has to be acknowledged that most of them are replaceable, and the words can be reprinted and reread. What is less replaceable is the intellectual world that produced them. Can artificial intelligence eventually become “intelligent” enough to participate meaningfully in literature, art, philosophy, and other disciplines which we have always proudly thought to be ours? After all, our essence is hardly measured by the artifacts we produce but rather the mortality, history, and relationships that produced them.
We may not be able to stop the development of technology, but we can still decide what type of relationship we want with the intelligence we are creating. The question, therefore, is not whether the machine will one day integrate itself into the things we once considered uniquely human. It is whether, by the time it does, we will still remember what made those things human to begin with.