In her “Venus in Two Acts,” Saidiya Hartman coins the phrase “critical fabulation.” Archives and historical records are filled with countless gaps and omissions, especially as it relates to the lives of enslaved people. As Hartman laments, the archive is “a death sentence,” and she wonders how she can do more than simply recount the litany of violence recorded in this figurative tomb.[1] In order to redress history’s omissions, Hartman uses storytelling to imagine not only what was, but also what could be. Starting with a single detail, Venus appearing as the placeholder name of a dead girl in a legal indictment against a slave ship captain tried for the murder of two girls, she imagines a narrative of friendship blossoming between the two girls in the hollow of the slave ship.[2] Hartman knows that such a counter-history may be seen as less legitimate, but a conventional history of the girls’ experience is impossible. Her semi-nonfictional, creative writing practice combines historical research, critical theory, and fictional narrative to address omissions in history. As she explains, these speculative arguments do not intend to falsify history so much as offer a critical reading of the archive.[3] For Hartman, poets, filmmakers and visual artists use fiction and its elasticity as a possible antidote to the violence that is a part of the everyday.

In my upper-level undergraduate class on Black media in Latin America—which introduces students to Black film and media produced in Brazil, Ecuador, Colombia, Venezuela, Peru, Cuba, Dominican Republic, Puerto Rico, and Mexico, with comparative hemispheric American reference to the United States—students read Hartman’s essay in a unit that prompts them to confront the stakes of historical narration. How does one recuperate lives entangled with archival utterances of banal brutality? Can narration “reanimate the dead” while respecting what cannot be known?[4] Can one narrate the past without aestheticizing violence? This contribution reflects on an assignment where we studied Black artists’ use of artifacts, archives, and testimonies in works that respond to the legacy of colonialism and its hold in the present.

For their capstone project, I asked students to attempt a “critical fabulation” starting with a single detail (a photograph, a testimony, an oral history, a recording, a mention or aside) from an archive about Black experience in the Americas. After their preliminary search in a digital archive, students developed a proposal during an in-class exercise with generative artificial intelligence (GenAI) to brainstorm and generate ideas for their fabulation. This in-class exercise yielded multiple opportunities for critical engagement with both the archive and artificial intelligence. Tavia Nyong’o builds on Hartman to propose we think of fabulation as a “dis/appearing act,” a metaphor that cannot help but recall the language of hallucination.[5]

What happens when fabulation and hallucination are thought in concert?

In Latin America, the question of race, nation, and multiculturalism is determined by the nation-building ideology of mestizaje—“the notion of racial and cultural mixture”—which became a project of national homogenization by way of the disavowal of racial difference.[6] In his influential study of race in Brazilian cinema, Robert Stam claims such a belief in national inclusion by way of racial mixture, a process ostensibly distinct from the racial segregation of the US context, resulted in “racial schizophrenia”: “the same Europeanized elite that conveniently invokes its mestiço culture... steadfastly refuses to empower its mestiço majority.”[7] In this light, the course examines how the meaning of race and ethnicity came to be defined through film and media and explores the media affordances of race in the region. I ask my students to consider how we might understand the legacy of racism differently in a region that incorporated and celebrated elements of Black cultural production as paradigmatic of national identity.

To answer this question, students engage with representations of race in film and media from Latin America. Salomé Skvirsky outlines the two dominant tendencies of representations of race in Latin American film and media: (a) historical narratives or period films, which she claims avoid reckoning with contemporary forms of racism by retreating to the colonial past and (b) national allegories, where the racialized subject becomes a symbol for the nation in an echo of the nation-building discourse of mestizaje.[8] Over the past five years, I have prompted students to write conventional analytical essays about the representation of race in these historical and allegorical films. Students would inevitably arrive at a progressive modeling of Black film historiography with a settled past of “problematic” representation and a future tied to a definitive sense of a victory achieved because of the more authentic representations of the reality of Black life.

In the last iteration of my class, confronted with both the growing use of GenAI in student writing as well as a growing dissatisfaction with these presentist arguments that reinforced representational binaries of positive and negative, I designed a new assignment that prompted students to reflect otherwise on historical narration. I suspected that a better way to have students reflect on historiography is by way of writing history, one that I coupled with an exercise in artificial intelligence.

Much of the apprehension of (or aversion to) GenAI in the writing-centered humanities classroom revolves around two predominant anxieties: concerns about (a) academic integrity and potential student misconduct and (b) academic veracity, the accurate transmission and reproducible creation of new knowledge. Taking a cue from preliminary research on GenAI use in the classroom that discourages restricting or eliminating writing assignments, I asked how I might use GenAI to meet the course objectives: Could GenAI present a new way for students to engage the way race is defined by film and media? More particularly, I wondered whether an assignment that used GenAI would allow students to reexamine the politics of representation in more self-reflexive ways and to unpack how GenAI produces and reinforces race anew.

I provided my students with a curated selection of digital archives to consult. Students found objects that piqued their interest and brought them to class to develop a proposal in conjunction with guided GenAI use in the classroom. Students were encouraged to use a GenAI platform available through the university, a tool that provides access to popular hosted AI models such as Azure OpenAI and open-source large language models (LLMs). I instructed students to use this platform to ensure access equity and data security, as the university does not share any user-specific data to improve AI models. Still, several students also used other GenAI tools such as ChatGPT, Bard (now Gemini), and Copilot.

I gave students four guidelines in preparation for their proposal: (1) analyze the archival artifact and consider its historic context to identify the questions it elicits; (2) lean into the ambiguities and uncertainties, linger on what you cannot ascertain or identify definitively; (3) consider counterfactual or alternative perspectives to speculate or diverge from the recorded fact; (4) incorporate metacognitive episodes to reflect on your position relative to the historical narrative and/or the limitations of the archive. Based on these guidelines, students formulated preliminary prompts. For instance, students attempted prompts to “imagine a world” shaped by the wake of their archival detail or “what if” their archival detail resulted in a different outcome. At a base level, the assignment provided a reflexive exercise in prompt engineering. Students prompted generative artificial intelligence with the material objects and primary sources they had culled from their preliminary archival research. In short, students prompted artificial intelligence to hallucinate. Although early uses of hallucination or confabulation in the field of artificial intelligence had more positive implications, referring to the process by which the computer produces a new image out of what it knows, the term now mostly refers to the generation of inaccurate or misleading information.

Students entered the first prompt they workshopped in class. I tasked students with reflecting on the first response artificial intelligence generated by honing their first prompt at least twice to improve their results and generate additional ideas for their creative project. Students found that their initial prompts were unspecific, lacking in context and ambiguous in language. In this way, they developed skills in prompt literacy as they iterated and refined their language, adding more context gleaned from the archive or providing clearer parameters for their “fabulation”. For instance, one student wanted to imagine a counter-history based on a photograph of a Haitian immigrant in Florida in the 1980s. An initial prompt, “imagine a day in the life of a Haitian immigrant” resulted in an unspecific and generic narrative. The student incorporated details from the image and from their own research about the historical context (i.e., escaping the Duvalier regime and settling in Little Haiti in Miami) to produce three ideas for their speculative historical narrative.

The bulk of discourse around artificial intelligence and its hallucinations dreads the tendency of Large Language Models to pass off fake or invented information as fact, to generate synthesized information that matches some pattern, but not reality. In this exercise, such hallucinations are leveraged productively to expand on Hartman’s call “to tell an impossible story and to amplify the impossibility of its telling.”[9] Students not only strained against the limits of the archive but also confronted how artificial intelligence conceives history as narrative data. Artificial intelligence scours the archive to retrieve information and trains neural networks on diverse datasets to generate responses based on learned patterns. Such pattern recognition at the heart of artificial intelligence, Wendy Chun reminds us, expands upon early work to discriminate between populations.[10] The use of GenAI to produce speculative historiography in this project enacts a second-order confrontation with the archive as dataset. What do we make of artificial intelligence trained on the “inventory” that Hartman encounters in the archive? What students found was that generative artificial intelligence simply produces more Venuses.

Prompting GenAI to hallucinate historical narratives about Black and Brown lives in the Americas resulted in fictions about historical figures that trafficked in romantic clichés. In the context of a class about the representation of race, students reflected on the historiographic prevalence of stereotype and how artificial intelligence reproduces the “innocent amusements” against which Hartman cautions. Hartman analyzes how melodrama presented blackness as a vehicle of protest and dissent by arousing pity and fear, desire and revulsion, and terror and pleasure. She argues that melodrama’s attempts to elicit empathy are double-edged: the white body must be positioned in the place of the Black body to make suffering visible and intelligible.[11] When prompting artificial intelligence to fabulate about slavery and its afterlives, students remarked that GenAI could not help but reproduce the melodramatic tropes in its dataset. Whether a historical fiction of enslaved persons endowed with magical powers or an immigration narrative centering on the white search and rescue officer’s encounter with Black and Brown death, the story ideas generated by GenAI seem to reinscribe subjugation and “reproduce essential and repressive definitions of blackness.”[12]

Armed with particulars from the archive, students also stumped artificial intelligence: for example, one student was told that her case study (i.e., Puerto Rican migrants in Hawaii) simply didn’t exist. The student used this error message to craft a speculative project with reflexive interludes which put her subjects’ vexed absence from datasets in tension with their stubborn if fragmentary presence in the archive. This assignment encouraged students to conceive the datasets that are fed to the computers as an archive and a basis of knowledge governed by power. In this way, the use of artificial intelligence’s hallucinations builds on the reflexive writing practice of critical fabulation, coupling a critical reading of the archive with a critical reading of LLMs and their models of generative information retrieval. Taking a cue from Trevor Paglen’s Adversarially Evolved Hallucinations (2017), which features machine-generated images fed with intentionally subjective categories of images to highlight the biases inherent in such systems, this assignment lingers in the powers of the false to illustrate the challenges of “narrating stories which are impossible to tell” in the age of hallucinations.[13] Paglen warns that AI’s generative modeling of the world estranges us from the present.[14] Could the more ethical relation to the past that critical fabulation enacts counteract such estrangement? If fabulation signals a “drive for the virtual,” a desire to project oneself in time and space, could we leverage the “powers of the false” to create new possibilities for the future?[15]

Could we encourage students to produce “adversarial” writing that exploits the limitations of datasets and the weaknesses of generative LLMs? This capstone project demonstrates how teaching writing with artificial intelligence requires perhaps departing from the conventional academic essay, opting instead for “adversarial writing” practices that are more self-reflexive and metacognitive. In this way, this project departs from most assignments with artificial intelligence that want to provide cautionary lessons about its inventions and falsities. In a way, what I’m suggesting is nothing new in the film and media studies classroom, which has long been a privileged site for laying bare the ways in which media shape and are shaped by ideology. We are tasked with renewing this project of demystification in the age of generative artificial intelligence.

Nilo Couret is an Associate Professor at the University of Michigan in the Department of Romance Languages and Literatures. His book, Mock Classicism: Latin American Film Comedy, 1930-1960 (University of California Press, 2018), traces the popularity and cultural significance of film comedies from the transition to sound through the industrial studio period. His articles have appeared in several edited anthologies and peer-reviewed journals, including Film History, Discourse, and Film Quarterly. 

1. Saidiya Hartman, “Venus in Two Acts,” Small Axe: A Caribbean Journal of Criticism 12, no. 2 (2008): 2, https://doi.org/10.1215/-12-2-1. ↑

2. Hartman, “Venus in Two Acts,” 1, 9. Venus does not represent an identifiable enslaved woman; instead, it stands in for the countless nameless Black women lost to history during the Atlantic slave trade. The name “Venus” was frequently given to enslaved women, highlighting their sexualization and commodification during that era. Venus marks their dis/appearance in the archive. ↑

3. Hartman, “Venus in Two Acts,” 12. ↑

4. Hartman, “Venus in Two Acts,” 3. ↑

5. Tavia Nyong’o, Afro-Fabulations: The Queer Drama of Black Life (New York: NYU Press, 2018), 17. ↑

6. Peter Wade, "Rethinking Mestizaje: Ideology and Lived Experience," Journal of Latin American Studies 37, no. 2 (2005): 239. ↑

7. Robert Stam, Tropical Multiculturalism: A Comparative History of Race in Brazilian Cinema and Culture (Durham, NC: Duke University Press, 1997), 48. ↑

8. Salomé Aguilera Skvirsky, “Las cargas de la representación: notas sobre la raza y la representación en el cine latinoamericano,” Hispanófila 177 (2016): 142–143, https://doi.org/10.1353/hsf.2016.0034. ↑

9. Hartman, “Venus in Two Acts,” 11. ↑

10. Wendy Hui Kyong Chun, Discriminating Data: Correlation, Neighborhoods, and the New Politics of Recognition (Cambridge, MA: MIT Press, 2021), 198. ↑

11. Saidiya Hartman, Scenes of Subjection: Terror, Slavery, and Self-making in Nineteenth-Century America (New York: Oxford University Press, 1997), 19. ↑

12. Hartman, Scenes of Subjection, 32. ↑

13. Hartman, “Venus in Two Acts,” 10. ↑

14. Trevor Paglen, Adversarially Evolved Hallucinations (London: Sternberg Press, 2024). ↑

15. Nyongo, Afro-Fabulations, 14, 67. ↑