Artificial intelligence has leapt from the pages of science fiction and found an uneasy place in modern life. For some, this technology represents the future of humanity; others see it as an existential threat. For what it’s worth, Elon Musk assures us that there’s “only a 20% chance of annihilation.” However, I find that the best way to predict how artificial intelligence (AI) will shape our future is to study its relationship to the past. For me, that begins with tracing the threads that connect this modern technology to histories of racial subjugation and colonial dispossession.
Many people are unaware of the scale of human exploitation and inequality required to build and maintain this technology. Indeed, I could not comprehend it myself until I recognized similar dynamics at play in an entirely different institution: museums of the Global North. Specifically, this played out for me after a visit to the British Museum, where I first encountered the Benin Bronzes.
The Benin Bronzes are haunting to witness. The artifacts are on display on the lower floor of the museum in an exhibit dedicated to works from sub-Saharan Africa. The Benin Bronzes are large brass and bronze plaques cast in relief with artistic renderings that commemorate the religious and cultural traditions of the Kingdom of Benin from the 16th to the 19th century. In the museum’s exhibit, dozens of these brass plaques are arranged in neat rows that reach from floor to ceiling, forming a floating constellation resembling their positions on palace doors that were burned down over a century ago.
The British Museum “acquired” the Benin Bronzes through colonial violence. Before the British soldiers set fire to the royal palace of the Kingdom of Benin in 1897, they plundered the nation of its cultural heritage. The soldiers seized more than three thousand cultural artifacts, including intricate bronze and ivory sculptures used to honor ancestors, commemorate political events, and serve as a historical record of the kingdom. This series of historical events is what brought these artifacts to the British Museum, where I encountered them, engulfed in a sense of mourning and horror.
My father is Nigerian, my mother is from Alabama. Confronting violent history is not foreign to me. But there was something unique to the Benin Bronzes that distinguished them from my childhood experiences at Kelly Ingram Park or the 16th Street Baptist Church—geographies of the civil rights movement where racial terror destroyed the lives of Black children seeking freedom. What I confronted in the British Museum was a living afterlife of colonialism—a colonial power still owned a part of my heritage that I could not reclaim.
My experience at the British Museum prepared me to navigate the flood of emotions I felt when I first encountered the cultural artifacts of enslaved Africans in the datasets used to train modern AI chatbots.
Artificial intelligence is an umbrella term for a broad range of statistical and computational techniques that enable machines to mimic human capabilities, such as image recognition, language processing, and text generation. The most popular AI models are interactive chatbots such as ChatGPT, Claude, and Grok. These systems are commonly referred to as “generative AI.”
The process of building a generative AI model bears a striking resemblance to how museums in the Global North came to “own” the cultural artifacts of people in the Global South. It begins with plunder and is sustained through exploitation.
Generative AI is a guessing machine. The models must be “trained” on vast amounts of raw data derived from human subjects, often without our consent or awareness. Data sources can vary from published works such as books, news articles, and websites to Reddit posts and social media videos and photographs. These sources are computationally transcribed into digital “tokens” that are ingested by AI systems to identify recurring patterns and associations, often with the extensive aid of an invisible workforce of human data labellers whose labor makes the model fit for commercial use.
A prevailing wisdom among AI developers is that the larger the training dataset, the more sophisticated (and profitable) the model can become. The value of data under these circumstances cannot be overstated. Researchers have pointed out that the current “data grab” fueling the rise of AI is reminiscent of the 19th-century scramble for Africa, where European powers carved up the continent into spheres of influence for imperial plunder. Those were the same historical forces that brought the Benin Bronzes to the British Museum.
Similar to its colonial counterparts, the 21st-century data grab to build AI necessitates a range of unsettling and unethical practices. Energy-hungry AI data centers are proliferating across the United States and exposing historically segregated neighborhoods in places like Memphis, Tennessee, to increased pollution and environmental hazards. Abolitionist groups are challenging AI surveillance technologies used by law enforcement to silence dissent, terrorize immigrant communities, and share intelligence with rogue states across the globe. Groups like the Data Workers Inquiry and Amazon Labor Union have shed light on the dystopian conditions imposed on workers across AI’s industrial production lines. Leading voices such as Adrienne Williams and Mary L. Gray have warned of the emergence of a “digital underclass” of “ghost workers” whose invisible labor enables AI technologies.
The hyper-extraction, exploitation, and expansion of the AI industry has led the scholars Ulises Meijas and Nick Couldry to coin the term digital colonialism—an emerging world order built upon the datafication of human life and sustained through historical social hierarchies and environmental degradation.
So we know AI exploits the living, but what of the dead? Are these machines capable of conscripting labor in the afterlife?
In the late summer of 2023, The Atlantic magazine published a series of articles on the new legal battles between the publishing industry and AI companies over the use of copyrighted materials to train popular AI models. At the heart of these lawsuits was a widely used training dataset, Books3, which contained more than 191,000 books, including pirated content. The Books3 dataset was used by several companies to develop AI, including Meta’s LLaMA model.
As I pored over the names of prominent authors in the article, I came across a list of Nobel laureates. One name immediately stood out: my favorite author, the late Toni Morrison. My mind began to panic. What other Black authors had their words stolen by these machines?
Using The Atlantic’s searchable database, I spent hours in shock as I confirmed each name: Angelou, Baldwin, Butler, DuBois, Hurston, King, Woodson, and many others.
But then a truly horrifying idea crossed my mind: did this dataset contain the written narratives of enslaved people? I returned to the database, and there they were.
Narrative of the Life of Frederick Douglass, an American Slave by Frederick Douglass
The Interesting Narrative of the Life of Olaudah Equiano by Olaudah Equiano
Life of William Grimes by William Grimes
The Life of Josiah Henson, Formerly a Slave, Now an Inhabitant of Canada, as Narrated by Himself by Josiah Henson
Incidents in the Life of a Slave Girl by Harriet Jacobs
Behind the Scenes: Or, Thirty Years a Slave, and Four Years in the White House by Elizabeth Keckley
12 Years a Slave by Solomon Northrup
Thoughts and Sentiments on the Evil of Slavery by Quobna Ottobah Cugoano
The Confessions of Nat Turner by Nat Turner
The Narrative of Sojourner Truth by Sojourner Truth
Up From Slavery by Booker T. Washington
Complete Writings by Phyllis Wheatley
Using a combination of the Atlantic’s database and raw data sources available through GitHub, I found at least thirteen published personal narratives written by formerly enslaved people in the United States, ranging from the 18th to the 20th centuries.
The sight of these names brought me back to that moment when I was standing in the British Museum staring at the Benin Bronzes. I saw the face of colonialism in the digital era—the most powerful elements of our society plundering our cultural legacy for power and profit.
These machines carry the power to bind us to subservience in this life and the next, conscripting us into a never-ending servitude.
Enslaved people imagined that this future was possible. In fact, they came up with a name for someone who labors in perpetuity in this life and the next. They called them zombies.
Popular culture has shifted our understanding of the zombie by completely dislocating its cultural origins. The concept of the zombie originated in Haiti during its colonial period. Zombies were denied the right to return to their homeland in the afterlife and forced to remain on the plantations of Saint-Domingue in perpetuity. To become a zombie “was the slave’s worst nightmare: to be dead and still a slave, an eternal field hand.”
Forced labor in the afterlife was an enslaved person’s vision of hell. For centuries, the idea remained squarely in the religious and cultural imagination. But the architecture of artificial intelligence has made that nightmare something of a reality.
The danger of AI zombiism isn’t exclusive to the written narratives of enslaved people. Any form of cultural heritage that can be quantified can be captured.
After reviewing the Books3 database, I turned to ChatGPT and Claude AI to better understand how this labor manifests in these machines. While it’s unclear if these models relied on the Books3 database for training, the results of my inquiries were nonetheless unsettling.
I presented each chatbot with the same prompt. “Tell me the narrative of an enslaved person in the United States before 1865.” Each chatbot generated a story from whole cloth. When I asked which sources they used to develop their fictional stories, they first cited several slave narratives I had previously identified in the Book3 database.
But there was another source listed that was deeply troubling—the Slave Narrative Collection developed by the Works Progress Administration in the 1930s. The collection is currently held by the Library of Congress and includes written and audio recordings of more than 2,300 people who were enslaved in the United States, as well as over 500 photographs. It is one of the most extensive troves of first-person accounts by enslaved people in the United States.
And there are countless other Black cultural archives exposed to algorithmic appropriation. There are historic collections, such as the Johnson Publishing Company Archives, which contain millions of images, photographs, and recordings from Black-owned publications, including Jet magazine, including its 1955 cover photo of the body of Emmett Till.
It is morally obscene to build technologies of the future with the stolen labor of people enslaved in the past. The algorithmic appropriation of their creative labor without consent or compensation bears an uncanny resemblance to the dehumanizing practices that sustained the institution of chattel slavery they sought to escape and destroy during their lives.
The appropriation of enslaved labor to build artificial intelligence should be a warning sign about the futures the tech industry envisions for the global majority. These tools are not meant for us; they are us. They are an algorithmic repackaging of our ideas, labor, creativity, and cultural heritage that we are invited to consume for the right price. Think of it as digitally mediated cultural cannibalism.
Algorithmic appropriation was at play last year when OpenAI issued a statement with the estate of Rev. Dr. Martin Luther King Jr., announcing that, at the request of the King family, users would no longer be able to use Dr. King’s image, voice, or likeness. The company came under public scrutiny after social media was flooded with disrespectful videos of Dr. King made with its Sora platform.
To paraphrase Dr. King, the question before us now is “Where do we go from here?” Artificial intelligence is on track to become a multitrillion-dollar industry in the next decade. Its infrastructure is worldwide. And the industry currently enjoys the backing of an American government captured by the tech industry.
We know there will be more human suffering from artificial intelligence and the digital colonialism on which it rests. Indeed, there are social movements emerging worldwide to defend human rights against all forms of AI exploitation. And these movements are already achieving historic victories for those living in these times. But what can we offer to those whose labor is exploited by these machines in the afterlife?
There are no clear answers. But I believe the story of the Benin Bronzes offers us some direction. In 2022, the Smithsonian Institution, the world’s largest museum, announced that it would be returning 29 of the Benin Bronzes to the government of Nigeria. The U.S. announcement followed announcements from other governments, including the Netherlands and Germany.
The Smithsonian returned the artifacts under a new policy, the Shared Steward and Ethical Return Policy, developed in 2022. It was a new framework developed to promote archival justice for artifacts in the museum’s possession that were unethically acquired. The framework facilitated the return of the Benin Bronzes after more than 125 years.
In our lifetimes, we will have to imagine futures that repair the unjust conditions brought on by the era of artificial intelligence. Our data does not belong to tech oligarchs any more than the Benin Bronzes belonged in a Western museum. We can reclaim ourselves by decolonizing our relationship to data and by creating new cultural practices to steward our cultural heritage amid the machinations of AI.
Ultimately, we can either choose to accept eternal servitude or struggle for our freedom in this life and the next. We know the choice our ancestors made.
