Social Media, AI, and the Racial Coding of Incompetence (Essay)
Content creator Matt Husk (also known as Husk IRL or @Husk) has built a successful TikTok comedy brand around an almost childishly simple premise: ask a large language model to do something easy and watch it fail. One of the clearest examples is his request for OpenAI’s ChatGPT to tell him how many e’s are in the word seventeen. The humor comes from the gap between the extraordinary claims made for artificial intelligence and the absurd incompetence of the machine in front of us. The system hesitates, then states confidently that there are two e’s in the word seventeen. Husk, meanwhile, remains visible on screen, reacting with disbelief to the spectacle. It is an effective comedic structure, and it is easy to see why he has found an audience.
What makes these videos more complicated, though, is the voice through which that failure is so often performed. Across much of Husk’s content, the AI assistant is given a voice that is recognizably Black-coded. That description needs some precision. A synthetic voice is not a Black person, and a voice that audiences perceive as Black does not necessarily speak African American English, or AAE. Race is being signaled through a collection of vocal qualities, speech patterns, and cultural associations that viewers have learned to read as racial. Yet that distinction does not make the coding insignificant. If anything, it makes the representational structure more revealing.
The recurring arrangement is striking. A white man occupies the visible, embodied position. He asks the obvious question. He knows the correct answer. He recognizes when the machine has failed. The Black-coded voice belongs to an unseen artificial entity that misunderstands instructions, speaks in circles, and cannot accomplish basic intellectual tasks. Whiteness is attached to a human body, authorship, judgment, and control. Blackness is reduced to a disembodied voice attached to confusion and incompetence.
This is where a seemingly innocent TikTok video becomes a visual-culture problem. Stuart Hall argued that representation does not simply reflect the world. It helps produce the categories through which the world becomes meaningful. A stereotype does not have to be announced explicitly to work. It gains power through repetition, familiarity, and association. In this case, no one has to say that Black people are unintelligent. The joke can remain entirely focused on the stupidity of artificial intelligence. But if the stupidity is repeatedly given a Black-coded voice while the rational observer is repeatedly embodied as white, an old racial hierarchy can be quietly restaged inside a new technological format.
That history matters because Black speech in the United States has long been treated as evidence of intellectual deficiency. African American English is a systematic variety of English with its own grammatical structures and histories, not broken or failed Standardized American English (Green, 2002). Nevertheless, listeners routinely attach judgments about education, intelligence, professionalism, and social status to the way people speak. A 2025 study by Nicole Gardner-Neblett, Angelica Ramos, and Allison De Marco found that White teachers judged children’s stories more harshly when those stories used features of AAE rather than Standardized American English, even when story quality was controlled. The bias was not simply about what the children said. It was connected to how their language was socially read.
Recent research on AI makes the situation even more uncomfortable. In a 2024 Nature study, Valentin Hofmann and his coauthors found that language models displayed strong covert prejudice toward speakers of AAE. The models could produce explicitly positive statements about African Americans while simultaneously associating AAE with negative traits and making worse simulated decisions about speakers who used it. The technology, in other words, has already absorbed many of the racialized assumptions embedded in the culture that produced it. When a Black-coded synthetic voice is then used as the comic vehicle for AI stupidity, the joke risks feeding those assumptions back to us as entertainment.
This does not require Husk to hold racist beliefs, and reducing the issue to a judgment about his personal character would miss the more important point. Images and sounds can participate in racist structures without being produced from a consciously racist intention. For a white viewer who already carries prejudicial assumptions about Black intelligence, the repeated pairing of Black-coded speech with spectacular incompetence can function as confirmation. It gives an existing stereotype another image, or in this case another voice, through which to feel natural. Because the racial content remains implicit, the association is also easy to deny. It is “just the AI.” That deniability may be precisely what allows the stereotype to circulate so comfortably.
Another side to the problem exists. People who are repeatedly subjected to negative representations of their own racial or cultural group do not absorb those images without consequence. The effect becomes persistent internalized racial hierarchy: the psychological burden of living within a culture that repeatedly attaches intelligence, beauty, authority, or competence to some identities and their opposites to others. Claude Steele and Joshua Aronson’s foundational work on stereotype threat demonstrated that the awareness of a negative stereotype about Black intellectual ability could itself affect performance under particular conditions. Representation is not a trivial background to that process. It helps furnish the stereotypes that people are forced to negotiate.
None of this means that Black-coded AI voices should disappear from comedy, or that every incompetent machine with such a voice constitutes a racist image. The issue is the pattern. What roles are repeatedly assigned to which voices? Who gets to be visible? Who gets to be human? Who possesses knowledge, and who performs failure? These are basic questions of visual studies because representation is built as much through relationships as through individual images.
Husk’s videos are interesting precisely because they are subtle and seem so inconsequential. They are short, funny, disposable pieces of social media built around the limitations of a new technology. Yet they also demonstrate how quickly artificial intelligence has become entangled with older systems of racial representation. Generative AI can produce voices without bodies, but those voices do not arrive without history. We hear race in them because culture has taught us how to hear it. Once that racial coding is paired repeatedly with stupidity, incompetence, or failure, the machine begins doing more than failing to count or misspelling words. It starts rehearsing a hierarchy that American culture has spent centuries teaching people to recognize.
Sources
Gardner-Neblett, Nicole, Angelica Ramos, and Allison De Marco. “When Ebony and Malik Share Stories in School: White Teachers’ Perceptions of Children’s Use of African American English During Oral Storytelling.” Early Childhood Research Quarterly 72 (2025): 143–155. https://doi.org/10.1016/j.ecresq.2025.02.006.
Green, Lisa J. African American English: A Linguistic Introduction. Cambridge University Press, 2002.
Hall, Stuart, ed. Representation: Cultural Representations and Signifying Practices. Sage/Open University, 1997.
Hofmann, Valentin, Pratyusha Ria Kalluri, Dan Jurafsky, and Sharese King. “AI Generates Covertly Racist Decisions About People Based on Their Dialect.” Nature 633 (2024): 147–154. https://doi.org/10.1038/s41586-024-07856-5.
Steele, Claude M., and Joshua Aronson. “Stereotype Threat and the Intellectual Test Performance of African Americans.” Journal of Personality and Social Psychology 69, no. 5 (1995): 797–811. https://doi.org/10.1037/0022-3514.69.5.797.
Murphy, Chris. “Comedians Are Trolling AI, and the Jokes Are Landing.” Vanity Fair, August 19, 2026.