PaperlayerView exact revision on arXiv

Source-linked AI summary

Shaping the Future of Generative AI for Black Communities: A Frame Analysis of Public Discourse and Empirical Scholarly Research

Angela D. R. Smith, Gabriella Thompson, Christopher L. Dancy, Mark Díaz, Seyi Olojo, Christina N. Harrington

arXiv:2608.24767v1cs.HCcs.CY

TL;DR

Research has not adequately connected public and scholarly accounts of genAI’s impacts on Black communities. This paper compares both corpora through Entman’s framing theory and finds structurally produced misalignment alongside a shared evacuation of Black epistemic agency.

  • Problem

    Public narratives and scholarly research shape understandings of genAI’s potential impacts on Black communities, but their framing and engagement require systematic examination.

  • Method

    The paper combines a systematic review of empirical research with media frame analysis, coding how each corpus defines problems, causes, and treatments.

  • Results

    Public discourse attributes genAI harm to historical and systemic forces, whereas scholarly research centers dataset-level causes and primarily technical reforms.

  • Takeaways & Limitations

    Frame analysis offers an AI ethics methodology for examining anticipatory and structural harms while centering Black communities as epistemic agents.

  • Takeaways & Limitations

    The media corpus represents narratives about Black communities rather than Black users’ and community members’ perspectives directly.

Abstract

from arXiv · show

As generative AI (genAI) systems become embedded in education, employment, healthcare, and creative industries, the impact and engagement among marginalized groups have become both a widespread discourse and a focus in scholarly research. As a starting point, we examine public discourse and empirical research to explore the impact of genAI systems on Black communities. We conducted a systematic literature review (SLR) of 91 empirical papers alongside a media discourse frame analysis of 28 public resources, applying Entman's framing theory to map how each corpus defines problems, attributes causes, and proposes treatments. Our SLR reveals that scholarly research concentrates heavily on technical bias detection, reducing Blackness to measurable variables rather than engaging with cultural practices, structural conditions, or Black knowledge systems. Our frame analysis reveals that public discourse attributes genAI-related harm to historical and systemic forces, while scholarly research stops its causal accounts at the dataset and its treatment recommendations at technical reform. We demonstrate that this misalignment is structurally produced: anti-Blackness operates simultaneously across both registers, generating a shared evacuation of Black epistemic agency. We argue for frame analysis as an AI ethics methodology capable of surfacing what technical evaluation forecloses.

Introduction

This paper examines how public discourse and empirical research frame genAI’s impacts on Black communities, identifying divergences between systemic accounts of harm and research focused on technical bias detection. It uses Entman’s framing theory and a systematic literature review to assess these priorities and their alignment.

  • Motivation: Public discourse shapes how emerging technologies are understood, adopted, and regulated as genAI becomes embedded across education, employment, healthcare, and creative industries.Media outlets, policy briefs, and other public narratives can inform research agendas, policy priorities, and community organizing.
  • Problem: The paper asks whether public framings of genAI and Black communities align with research priorities studying and engaging those communities.It also interrogates how Blackness, culture, and identity are conceived, engaged, and portrayed in AI development.
  • Method: The study combines Entman’s framing theory with a systematic literature review of empirical research to compare public narratives and scholarly priorities.The frame analysis examines media resources for problem definitions, causal attributions, moral judgments, and proposed treatments, while the review characterizes research from computing venues.
  • Contribution: Public discourse and empirical scholarly research diverge, with public frames emphasizing systemic bias and marginalization while scholarship concentrates heavily on technical bias detection.The paper identifies and characterizes dominant public frames concerning genAI’s anticipated impacts on Black communities.
  • Contribution: Public frames reveal anticipated harms, resonant causal explanations, and viable interventions that can inform AI ethics research agendas and policy development.Mapping public concern alongside research focus supports efforts to ensure AI research addresses these issues.

Blackness, Racialization & Relation to GenAI

Understanding genAI’s relationship with Black communities requires situating contemporary bias within historical racialization and examining how anti-Blackness enters AI development across levels. In HCI, solutionist and colorblind approaches can treat Blackness as a problem while obscuring structural racial logics and the complexity of Black experience and knowledge.

  • Historical and Structural Context: Contemporary genAI bias is shaped by racial structures influencing how AI systems are developed, deployed, and experienced at individual, interpersonal, organizational, and institutional levels.The passage locates these influences across micro- and meso-level processes.
  • Historical and Structural Context: Historical racialized logics shifted from theological to biological to cultural rationales while consistently positioning Blackness as inferior or deviant.These logics operate through “inlets” through which anti-Blackness enters AI development at every level.
  • HCI and Racialized Framing: HCI solutionist approaches treat Blackness as a problem to correct rather than recognizing the ontological complexity of Black experience and knowledge.This framing reduces Blackness to an object of technical correction instead of engaging its complexity.
  • HCI and Racialized Framing: Colorblind perspectives and fluid racial categories obscure the persistence of racialized logics, making structural intervention difficult.The passage identifies both perspectives and category fluidity as barriers to recognizing ongoing structural effects.

Related Work

Prior research shows that AI development prioritizes quantitative efficiency and performance, while documenting sociotechnical harms and emphasizing that Black communities hold complex, often critical views of AI. This work has prompted calls for critical, reflexive, and participatory approaches to AI development.

  • Values and harms: AI scholarship prioritizes quantitative empiricism, efficiency, and performance over societal need and negative impacts, reflected in benchmark datasets valuing efficiency and universality over care and contextuality.These priorities shape how AI systems are built and assessed, with consequences for minoritized communities.
  • Values and harms: Research documents racial bias in language models, gender bias in image generation, and frameworks for identifying and mitigating model bias.Shelby et al.’s taxonomy organizes sociotechnical harms into representational, allocative, quality of service, interpersonal, and social system categories.
  • Black perspectives and participation: Black communities hold complex, often critical attitudes toward AI, particularly in healthcare contexts, and report pessimism about algorithmic fairness and bias in online platforms.Public perceptions of AI vary across race, class, and gender and influence how systems are developed and deployed.
  • Black perspectives and participation: Scholars call for responsible inclusion of Black communities through feminist-reflexive data curation, critical race theory for AI fairness, CRT-grounded HCI, and participatory co-design.These approaches seek to incorporate Black communities into AI development and have demonstrated particular value in co-design with Black adults.

Methods

The study combines a systematic literature review of empirical research with an Entman-based frame analysis of public discourse to identify misalignments between scholarly priorities and community-identified concerns. The review yielded 91 coded papers, while the public corpus was analyzed for framing functions and treatment recommendations.

  • Study design: The researchers combined a systematic literature review with public-media frame analysis to compare empirical research priorities with public concerns about genAI and Black communities.The comparison was intended to reveal research misalignments and potential failures to address community-identified needs and anticipatory harms.
  • Systematic literature review: The review searched major computing and AI databases for full, peer-reviewed, English-language empirical papers published from 2010–2025 that focused on Black populations and genAI.Eligible studies addressed genAI involving images, text, and/or video and qualitative engagement such as interviews, focus groups, or participant research.
  • Systematic literature review: 223 papers were retrieved, 71 were removed as irrelevant, 61 as duplicates, and 18 for focusing on marginalized groups broadly, leaving 91 papers for coding.The final corpus included 19 IEEE, 41 ArXiv, 13 ACM, and 18 ACL papers.
  • Coding and analysis: The coding instrument extracted publication, venue, study, genAI, contribution, application, artifact or dataset, evaluation, engagement, and Blackness-related dimensions from the 91 papers.All authors iterated on the instrument, tested it on five papers for reliability and validity, and then divided the remaining corpus among four researchers.
  • Frame analysis: The frame analysis used 28 public sources and Entman’s four functions: problem definition, causal interpretation, moral evaluation, and treatment recommendation.Treatment recommendations were coded as strategies or calls to action intended to avoid or alleviate genAI’s negative effects.

Findings · Systematic Literature Review Results

The 91-paper review found rapid recent growth, concentrated attention to racial and gender bias, and a dominant emphasis on model evaluation and benchmark development. Black communities were usually treated as subsets or comparison groups, while most studies framed bias as a disproportionate harm and pursued testing or stereotype reduction.

  • Findings: The findings report descriptive results from the 91-paper corpus and interpret patterns through problem definition, causal interpretation, and treatment.The analysis places systematic-review findings and public-discourse findings in conversation using an onto-epistemological framework.
  • Trends in Types of Empirical Research: 54 papers addressed racial and gender bias in text-to-text and text-to-image models, while smaller groups addressed Economics, Health, and other application domains.The passage specifies three Economics papers and ten Health papers, with additional domain counts truncated in the supplied text.
  • Methods and Technological Focus: 52 papers were model evaluations, 25 developed benchmarks or datasets, and 12 developed artifacts or systems.Four dataset-development papers used surveys, and four dataset-evaluation papers reported human annotation involving a subset of Black communities.
  • Methods and Technological Focus: 18 papers focused on text-to-text models, 18 on text-to-image models, 16 on named models or systems, four on text-to-media, and one on image-to-image.Examples of named systems included ChatGPT, StyleGAN2, and visual language models.
  • Communities of Focus and Relationality to Blackness: 28 papers focused solely on Black and/or African American communities, whereas 38 treated Black communities as subsets or comparisons with other racial groups.Some papers advocated participatory engagement, co-learning, empowerment, or community-centered archival practices without directly engaging Black community members.
  • Trends in Research Contributions: Most papers framed bias as a potential harm disproportionately affecting Black or darker-skinned communities, including studies of stereotype replication and stereotype classifiers.Research contributions commonly involved testing genAI models or investigating stereotypes, norms, and identity narratives across models.

Framing Results

The 28 media resources consistently attributed genAI-related harm to systemic and historical bias, while treatment recommendations most often emphasized regulation and AI-development reform. However, two recent sources on data-center infrastructure and environmental harms identified problems without proposing interventions.

  • Causal attribution: 25 of 28 media sources identified systemic bias as a causal force, while 21 of 28 identified historical bias.More than half cited both simultaneously, treating systemic and historical bias as structurally intertwined explanations.
  • Treatment recommendations: 13 of 28 media resources recommended regulation, followed closely by 12 of 28 recommending reform of AI development.These treatment recommendations reveal the corpus’s structural orientation.
  • Treatment recommendations: Two recent sources on data-center infrastructure and environmental harms to Black communities offered no treatment recommendations.They contributed substantially to the prevalence of social system harm coding, identifying problems without proposing interventions.

Problem Definition: Representational Harm as a Shared Epistemic Frame

Representational harm is the only problem definition appearing with comparable force across media and scholarly corpora, making it the analysis’s organizing frame. Both corpora document harms to Black representation while omitting Black communities as knowledge producers, technological leaders, and epistemic agents.

  • Shared Frame: Representational harm is the only problem definition appearing with comparable force across both corpora.Social system harm was most frequent in the media corpus, but representational harm remained the organizing frame because it was shared across both corpora.
  • Media Corpus: Media accounts described genAI outputs that erased, distorted, or stereotyped Black identity.Examples included progressively lightened images of darker-skinned subjects, stereotypical depictions of Black Americans, and the erasure of Black cultural content.
  • Media Corpus: Black students were more than twice as likely as white or Latino students to be falsely accused of using AI to write their work.These interpersonal harms received no parallel treatment in the scholarly corpus.
  • SLR Corpus: Scholarly research likewise identified representational harm most frequently, with stereotyping as its most common descriptor.Studies examined stereotyping in LLM personas, homogeneity in representations of darker-skinned Black individuals, and amplified race and gender stereotypes in image generation.
  • Shared Absence: Neither corpus centered Black communities as knowledge producers, technological leaders, or epistemic agents in genAI development.No media frame centered Black-led innovation, community-driven design, or Black technological agency as a structural response to representational harm; the research corpus included 91 papers, 14 engaging Black communities directly.

Causal Interpretation Across Both Corpora

The media corpus explained genAI-related harms to Black communities through historical and systemic forces, whereas empirical research often removed causal explanation or located bias in training data. This divergence reproduces an epistemic tendency to avoid addressing Blackness and anti-Blackness directly.

  • Causal Interpretation Across Both Corpora: Media discourse consistently attributed genAI-related harms to historical and systemic forces, while empirical research exhibited a pattern of structural evacuation.The empirical corpus removed causal explanation from research accounts in ways that reproduce an epistemic tendency to skip over Blackness and anti-Blackness directly.
  • Causal Interpretation Across Both Corpora: 25 of 28 resources identified systemic bias as a causal force linked to current policies, organizational cultures, and other structural conditions.The passage lists systemic bias as a pervasive and structurally oriented causal attribution in the media corpus.
  • Causal Interpretation Across Both Corpora: 30 of 91 papers included no causal interpretation, treating AI systems instrumentally without identifying forces responsible for reported harms or limitations.These papers evaluated or sought to improve AI systems without implicating broader causal forces.
  • Causal Interpretation Across Both Corpora: 55 of 91 papers attributed bias to embedded bias in training datasets, making dataset-level mechanisms the most common causal account among papers offering one.The SLR corpus therefore diverged sharply from the media corpus’s historical and systemic causal framing.

Treatment Recommendation: Technical Reform, Structural Protection, and the Shared Evacuation of Black Epistemic Agency

Treatment recommendations followed the scope of each corpus’s causal account: dataset-level explanations produced technical reform, while structural and historical explanations reached toward regulation and systemic intervention. Yet both corpora positioned Black communities as recipients of action rather than epistemological subjects determining AI’s conditions.

  • Treatment Recommendation: Dataset-truncated causal accounts concentrated treatment recommendations on technical reform, whereas structurally extended accounts reached toward regulation and systemic intervention.The recommendations followed directly from whether causal interpretation stopped at the dataset or implicated the structural conditions producing the data.
  • Technical Reform: 78 of 91 SLR papers recommended AI-development improvements as their primary treatment, including better datasets, revised annotation, improved benchmarks, and algorithmic bias reduction.No empirical paper recommended regulatory intervention, community-led governance, or structural changes to AI-development conditions.
  • Structural Protection: 13 of 28 media resources called for regulatory intervention involving legislators, educators, and institutional policymakers governing genAI’s impact on Black communities.These recommendations addressed agents capable of governing genAI’s effects on Black communities.
  • Structural Protection: 9 of 28 media resources recommended participatory strategies to prevent or mitigate harm, while economic media urged employers to adopt participatory AI planning.The media corpus also commonly recommended reforming AI development itself.
  • Shared Evacuation of Black Epistemic Agency: Across treatments including access expansion, regulation, resistance, and reform, both corpora positioned Black communities as recipients rather than epistemological subjects determining AI’s conditions.Protective action was assigned to legislators, educators, developers, and employers; resistance was framed as individual refusal rather than collective agency.

Discussion

The discussion argues that misalignment between public discourse and scholarship on genAI and Black communities is structurally produced through an epistemological enclosure that limits Blackness to harm and technical bias. It calls for research and development that recognize Black knowledge, agency, joy, freedom, and world-making capacity while acknowledging limitations in whose perspectives media discourse represents.

  • Structural misalignment: Misalignment is structurally produced: public discourse attributes genAI harm to historical and systemic forces, while scholarship stops at dataset causes and technical reform.Public discourse reaches toward regulatory and structural interventions, whereas the SLR corpus primarily recommends technical reform.
  • Limitations: Media discourse does not straightforwardly represent Black communities’ perspectives, because journalistic news values and editorial priorities shape its narratives.The analysis characterizes public narratives about Black communities and genAI rather than directly capturing Black users’ and community members’ views.
  • Epistemological enclosure: Media and empirical research jointly enclose Blackness within racialized boundaries that position Black people, communities, and practices as objects, deviant, and lacking.The discussion characterizes this as an epistemological enclosure of Blackness.
  • Epistemological enclosure: Both corpora systematically foreclose Blackness as a generative, world-making capacity, making this absence a structural condition rather than a minor oversight.Blackness is presented as a site of knowledge, agency, and ontological possibility, not only harm.
  • Liberatory futures: Liberatory genAI futures require interrogating and conditionally accepting these tools while cultivating belonging and freedom through participatory, considerate approaches.The discussion also emphasizes holding space for existing Black joy, play, and freedom.

Conclusion

The study finds a structural misalignment between public discourse, which emphasizes historical and structural harms, and scholarly research, which offers limited causal explanation and favors technical reform. It also advances frame analysis as an AI ethics methodology for anticipatory, structural questions.

  • Public discourse emphasizes historical and structural dimensions of genAI-related harm as important causal forces affecting Black communities.
  • Scholarly research often lacks causal explanation or relies on shallow, technical, infrastructure-oriented accounts while focusing on technical, bias-oriented reform.
  • Frame analysis is proposed as an AI ethics methodology suited to anticipatory and structural questions.
  • Positive effects of public discourse on scholarly research are not guaranteed because research can be epistemologically structured to exclude them.

Positionality Statement

The six-member research team’s intersecting identities, lived experiences with anti-Blackness, and expertise inform its analytic commitments, while institutional proximity to genAI development both affords access and constrains the work. The team holds this tension explicitly rather than resolving it, allowing the analysis to follow evidence across both corpora.

  • Research team: The six-member team spans graduate, faculty, and industry research roles, with most members of African and African American descent and expertise in critical HCI, participatory research, and anti-Blackness-centered AI.Their lived experience with anti-Blackness informs, but does not exhaust, their analytic commitments.
  • Institutional positionality: The team’s academic and industry positions provide access and insight while constraining what is easy to say, because its industry organization’s parent company develops and deploys the genAI systems it critiques.The researchers explicitly hold this tension rather than resolve it and let the analysis follow evidence in both corpora.
Loading 2608.24767v1…