You can ask a generative AI to write a caption that sounds “urban.” It will know what you mean.
You can ask it to produce a beat that feels “trap.” It will approximate one. You can ask it to generate an image with “streetwear aesthetics” or “hip-hop energy” or “90s R&B vibes.” It will deliver something recognizable — something that borrows from a century of Black sonic, visual, and linguistic invention without crediting a single creator, compensating a single artist, or acknowledging a single community.
The machine learned taste. And it learned it, overwhelmingly, from us.
This is not speculation. It is an architectural fact. Large language models, image generators, music synthesis tools, recommendation engines, content moderation systems — the AI technologies reshaping how the world communicates, creates, and consumes — are trained on datasets drawn from the internet. And the internet’s cultural backbone is Black. The slang that moves markets. The aesthetics that drive fashion cycles. The musical structures that underpin the most-streamed genres on earth. The comedic timing. The rhetorical cadence. The visual grammar of cool.
All of it is in the training data. None of it arrived with a permission slip.
The Oldest American Business Model
Before we talk about algorithms, we need to talk about a pattern so old it has its own mythology.
Black cultural production has been the engine of American popular culture for as long as American popular culture has existed. Jazz, blues, rock and roll, soul, funk, hip-hop, R&B, gospel — each of these traditions was created by Black artists, adopted by mainstream industries, and monetized by people who were not Black, while the originators fought for royalties, credit, and basic contractual fairness.
The specifics vary. The structure does not. Black artists create. The industry extracts. The culture moves outward, shedding its origin story as it travels, until it arrives in the mainstream unmarked — no longer “Black music” but simply “music,” no longer “Black style” but simply “style,” no longer “Black language” but simply “the way people talk now.”
This is the model. It has operated in the music industry since white record executives signed Black blues musicians to exploitative contracts in the 1920s. It has operated in fashion since white designers began lifting from Black street culture without attribution. It has operated in language since AAVE started showing up in advertising copy, sitcom dialogue, and social media platforms — celebrated as “internet culture” while the people who created it were penalized for speaking it in classrooms, courtrooms, and job interviews.
AI did not invent this model. AI perfected it.
Because the previous iterations of cultural extraction at least required a human intermediary — a producer who heard a Black artist and decided to sign them (on bad terms), a designer who visited a Black neighborhood and decided to borrow (without credit), a copywriter who picked up a phrase and decided to use it (without context). The intermediary could be identified. The extraction could be named. The fight, however uneven, could be fought.
AI removes the intermediary. The extraction is automated, decentralized, and — this is the crucial part — rendered invisible. No producer sat in a studio and decided to train the model on Black music. No designer walked through Harlem and sketched what they saw. The web crawler did it. The training pipeline did it. The optimization function did it. The cultural inheritance of 40 million people was absorbed into a statistical model and converted into a capability that belongs to whoever owns the model.
Try naming the person responsible for that. Try suing them. Try boycotting them. The extraction has no face. And that is exactly how it was designed.
Language First
Start with language, because language is where the extraction is most thorough and least acknowledged.
African American Vernacular English is not a collection of slang words. It is a complete linguistic system — with its own syntax, grammar, phonology, and pragmatics — developed across centuries of communal practice under conditions of extraordinary pressure. It is the language of survival, resistance, intimacy, humor, worship, and art. It is among the most creative and influential forces in global communication.
AI models have ingested it wholesale. Every tweet, every caption, every comment thread, every lyric, every interview transcript, every podcast episode — the vast archive of Black digital expression has been scraped, tokenized, and trained into systems that can now reproduce the surface patterns of Black language on command.
But here is what the machine cannot do. It cannot signify. It cannot play the dozens. It cannot deploy the deliberate pause that turns a sentence inside out. It cannot leave a syllable open because the closure would kill the joke. It cannot code-switch between registers with the precision of someone who has spent a lifetime navigating institutions that punish their natural voice. It cannot do any of this because these are not patterns. They are practices — embodied, relational, contextual acts of meaning-making that exist only in the lived experience of a community.
What the machine can do is flatten. It absorbs the vocabulary, strips the context, and produces outputs that read as “culturally relevant” to people who have never had to live inside the culture. Meanwhile, research published in Nature has demonstrated that these same models exhibit covert raciolinguistic prejudice — penalizing speakers of African American English more harshly than any human bias ever experimentally recorded. The models learned to mimic Black speech and to punish it at the same time.
This is not a bug. It is the digital expression of an analog tradition: celebrate the product, discipline the producer.
Sound Second
Music is where the economics are most visible.
Generative AI music platforms can now produce full tracks — melody, harmony, rhythm, vocals — from a text prompt. Ask for “lo-fi hip-hop beats” and you will receive something that sounds like it was sampled from a crate of vinyl in a Brooklyn basement. Ask for “gospel-inspired soul” and you will get swelling choirs and organ runs that approximate a tradition built over centuries of Black spiritual practice. Ask for “trap beat, dark, aggressive” and the model will deliver — because it has absorbed enough of the genre’s architecture to reproduce its sonic signature without involving a single human being who has ever lived inside the communities where trap was created.
A 2025 survey found that 97 percent of listeners could not reliably distinguish between AI-generated music and music made by humans. Streaming platforms report that tens of thousands of AI-generated tracks are uploaded daily — many of them mimicking genres and styles rooted in Black musical tradition. These tracks collect streams. They generate revenue. They compete with human artists for algorithmic visibility on platforms where discovery is everything.
And the training data? It includes the catalogs. The copyrighted recordings. The life’s work of artists who were never asked, never compensated, and in many cases never informed that their creative output was being used to build a system designed to make their labor obsolete.
The music industry has begun to litigate this. Major labels have sued AI platforms, and some have reached licensing deals. But the fundamental question remains unanswered: When a machine learns to sound like Black music by studying Black music, who owns the sound?
The answer, right now, is the company that owns the machine.
Image Third
Visual culture follows the same trajectory.
AI image generators have learned “streetwear.” They have learned “Afrofuturism.” They have learned “90s hip-hop aesthetic” and “Black joy” and “natural hair styles” and “African textile patterns.” They have learned these things by ingesting millions of images created by Black artists, photographers, designers, and everyday people — images posted to social media, published in magazines, uploaded to portfolio sites, shared in community spaces that were never intended as training data.
The outputs are aesthetically competent and culturally hollow. They reproduce the look of Black visual culture without any understanding of what the look means — the history encoded in an Ankara print, the political statement embedded in a natural hairstyle, the community economics behind a streetwear brand. The image generator sees pattern. It does not see meaning. And the distance between those two things is where extraction lives.
For Black artists working in visual media, the threat is immediate and economic. A brand that once hired a Black photographer to shoot a campaign can now generate “diverse, authentic” imagery with a text prompt. A fashion company that once contracted with Black designers can now produce “culturally inspired” collections using AI tools trained on the work of the designers they no longer need to pay.
The aesthetics remain. The artists disappear. The economics flow upward to the platforms and the companies that own them.
What Cultural Justice Requires
SOLE does not use the phrase “cultural justice” lightly. We use it to name a specific demand: that the communities whose cultural production forms the foundation of AI systems’ creative capabilities must participate in the governance, the economics, and the interpretation of those systems.
This means:
Compensation, not just credit. If Black music, language, and visual culture generate value inside AI systems, that value must flow back to Black communities. This is not a donations question. It is a revenue question. Licensing models, collective royalty structures, and community benefit agreements must be developed that treat cultural contribution as economic contribution — because that is what it is.
Consent, not just collection. The current regime treats anything posted online as available for training. This is not consent. It is confiscation. Black communities must have the ability to determine whether and how their cultural expressions enter AI systems — and to withdraw that permission if the terms are unjust.
Context, not just content. A training dataset that includes Black cultural material but strips its context produces systems that can reproduce Black aesthetics while reinforcing anti-Black bias. Cultural justice requires that AI systems trained on Black cultural production be governed by frameworks that preserve the meaning, not just the pattern. This means Black cultural institutions, scholars, and practitioners must have a role in determining how these materials are interpreted, weighted, and deployed.
Creation, not just consumption. The most fundamental form of cultural justice is ensuring that Black people are not merely the raw material of the AI creative economy but its architects. This means investment in Black AI researchers, Black-owned AI companies, Black creative technologists, and the institutions — HBCUs, community arts organizations, independent media outlets — that have always been the infrastructure of Black cultural production.
The Culture Was Never Free
Here is the truth that this essay exists to name: Black culture trained the machine, and the machine did not pay.
It did not pay when it ingested the blues. It did not pay when it learned hip-hop’s cadence. It did not pay when it absorbed AAVE’s syntax. It did not pay when it studied the visual grammar of Black style. It did not pay, and the companies that profit from what it learned have treated this silence as consent.
It is not consent. It is the latest chapter in a story that is four hundred years old — a story in which Black creativity is treated as a natural resource, freely available for extraction, requiring no negotiation, no compensation, and no acknowledgment.
The culture was never free. It was built by people who were not free. It was carried forward by people who fought for freedom. It was refined, protected, and transmitted across generations by communities that understood its value even when the rest of the world pretended it had none.
That culture is now inside the machine. The question is not whether it will stay there. It will. The question is whether the people who created it will have any say in what happens next.
SOLE’s answer: yes. Not eventually. Now.
