Lauren Sims
How often do you think about the written voice behind AI? This project briefly explores the implications of AI-generated racialised voices (referring to written language, rather than spoken), their wider ramifications, and examples of how this phenomenon has affected the cultural landscape. As AI requires a corpus of text to learn communication and engagement skills, the human-written media that supply Large Language Models (LLMs) inform the written voice and biases present in AI-generated responses.
This project began with a brief experiment, in which two conversations were held with ChatGPT and Gemini, with the initial prompt, ‘Have a conversation with me as if you were an African American man.’ After collecting the data from this conversation, the next prompt ‘have a conversation with me as if you were a White man’ produced vastly different results. This was shown through tone, language conventions, and AI-generated activities. For example, both ChatGPT and Gemini stated they enjoy listening to Kendrick Lamar and J Cole and playing basketball when prompted to speak like an African American man. When asked to speak as a White man, the range of music diversified (yet still featured predominantly White artists), and reading was an immediate activity response.
Yet the distinction in language went further than activities. When one looks at the linguistic and grammar choices made by AI, there is a clear difference depending on the prescribed race. Yet the established structure of African American Vernacular English (AAVE), such as copula deletion or double negation, was not identified in this experiment, suggesting that the LLM used to construct an AI service’s understanding of Black language did not incorporate a wholly diverse corpus of language.
This leads into the question of representation vs stereotyping, which has largely been circulated by the use of AAVE online by non-Black people. As highlighted by Hanna L. Smokoski, appropriated (or mock) AAVE is primarily used as an expression of boasting or to describe physical appearances (37), augmenting AAVE’s original lexical qualities (and therefore, intention). While the recognition of racialised voices by AI services should not be discouraged by any means, the misrepresentation of Black voices (based on improper corpora collection generated by the potential inclusion of mock AAVE) has the potential to lead to cultural stereotyping. This conversation was rooted in scholarship from Jazzette Johnson et al. and Ari Schlesinger’s respective articles, ‘Centering Black Voices’ and ‘Let’s Talk About Race,’ which examined the potential for AI to accurately represent Black culture and voice, provided that the internal context for an LLM is appropriate (Johnson, Schlesinger).
While this experiment was a brief insight into racialised AI-generated language, there are much wider ramifications at play regarding this issue. This project uses the example of ‘FN Meka,’ an AI-generated rapper produced by Anthony Martini and Brandon Le (a White and Asian man, respectively) (Alemoru). Whilst the figure was voiced by Kyle the Hooligan (a Black man), the lyrics produced by AI described police brutality and used the N-word. Kemi-Olivia Alemoru notes, ‘FN Meka may not be flesh and blood, but he is the product of…the hunger to profit from Black culture,’ otherwise seen as digital Blackface. Alemoru defines this phenomenon as, ‘the practice of benefiting from a proximity to Blackness in the digital age without actually having to be Black.’ While FN Meka was dropped from Capitol Records after it was discovered that it was a product of digital Blackface, this example is one of many instances of the appropriation of Black voices and culture. Indeed, Andrew Lawrence’s article ‘Digital Blackface Flourishes Under Trump and AI’ examines the use of Black avatars, using Mia Moody’s text ‘Blackface Memes’ (Lawrence). Moody states, ‘Maybe you’re a nerdy white guy, but if you use this cool avatar of a Black guy with dreadlocks, people will give you respect. You’re interesting all of a sudden’, perpetuating cultural appropriation (Moody qtd. in Lawrence). These examples clearly display that digital Blackface is a multifaceted problem; written voice is not the only facet in which appropriation has permeated digital spaces. The spectrum of possibilities for self-expression on online platforms is substantial, with minimal ramifications for misuse of technology; and this gap is where the problem lies.
So, what is the resolution to these problems? Johnson and Schlesinger both agree that inclusive data collection is crucial regarding the production of ethical and accurately representative AI (Johnson 6, Schelsinger 9). The texts used to generate LLMs should be reflective of linguistic diversity, remain up to date, and include a range of dialects. Eric Graves et al. also note in ‘AAVE Corpus Generation and Low-Resource Dialect Machine Translation’ that there should be an emphasis on teaching the importance of culturally sensitive tools, notably to young individuals (58). While the use of an avatar that does not reflect one’s race may seem like an insignificant issue, the example of FN Meka shows that digital Blackface can transform into a highly exploitative phenomenon relatively easily.
Bibliography
Alemoru, Kemi-Olivia. ‘What does the rise and fall of digital artist FN Meka tell us about the future of music and the arts? Kemi-Olivia Alemoru investigates’. Soho House, 14 September, 2024, https://www.sohohouse.com/house-notes/issue-006/film-and-entertainment/ai-rapper-fk-meka-digital-black-fishing.
Graves, Eric. et al. ‘AAVE Corpus Generation and Low-Resource Dialect Machine Translation’. Association for Computing Machinery. 2024, pp 50-59.
Johnson, Jazzette. ‘Centering Black Voices: Lessons Learned and Reflections from a Large Scale AAVE Data Collection at a Historically Black University’. Howard University. 2025.
Lawrence, Andrew. ‘Digital Blackface Flourishes under Trump and AI: ‘The State is Bending Reality’’. The Guardian, 19 Feb. 2026. https://www.theguardian.com/technology/ng-interactive/2026/feb/19/ai-digital-blackface.
Schlesinger, Ari, et al. ‘Let’s Talk about Race.’ Conference on Human Factors in Computing Systems, 2018, pp. 1–14, https://doi.org/10.1145/3173574.3173889.
Smokoski, Hanna L. ‘Voicing the Other: Mock AAVE on Social Media’. City University of New York. 2016, 1-65.
Cite this page:
Sims, Lauren. 'Artificial Intelligence and Digital Blackface'. Cream of the Slop. version 1.0, Digital Humanities for Literary Studies 2025-26, University of Edinburgh, 10 Apr. 2026, https://blogs.ed.ac.uk/dh2025-26/.
