AI systems designed without valuing frontline worker expertise consistently fail and deepen the devaluation of feminized labor.

Objective:
The primary goal of this study is to examine how the systemic devaluation of worker expertise—particularly in feminized professions such as K–12 teaching, home healthcare, and social work—shapes the repeated failures of workplace AI systems. The authors aim to introduce and explain the concept of AI Failure Loops, cyclic patterns in which flawed AI design reinforces occupational devaluation, ultimately undermining both workers and the systems intended to support them.

Methods:
The study uses a focused literature review across academic and grey sources, synthesizing decades of research from human–computer interaction (HCI), science and technology studies (STS), organizational science, and labor sociology. The authors analyze AI deployments in three feminized labor contexts: K–12 education, home healthcare, and social work. Their method involves mapping patterns of AI design, evaluation, and deployment failures to broader sociotechnical conditions, especially the historical and structural undervaluation of feminized labor. Through comparative analysis, the authors highlight recurring themes such as lack of worker involvement, reductive representations of labor, oversight of tacit knowledge, and political dynamics in technology adoption.

Key Findings:

• AI systems repeatedly fail in feminized labor sectors because developers underestimate the complexity, nuance, and tacit expertise embedded in frontline work.

• Workers—especially women and people of color—have historically been excluded from decision-making roles, leading to AI design processes that do not reflect real work practices or needs.

• Devaluation of expertise manifests structurally (low pay, low occupational status) and culturally (misconceptions about emotional labor as “acts of love”).

• These conditions create AI Failure Loops:

  • Worker expertise is ignored → flawed AI tool is designed → AI deployment fails in practice → failure reinforces beliefs that workers are the problem rather than the system → further devaluation of labor → next AI tool replicates the same issues.

• Across all three domains, AI systems often increase workload, compromise autonomy, or introduce bias rather than delivering promised efficiencies.

• Even “participatory design” efforts fall short because workers lack actual power in the design process, producing superficial engagement rather than meaningful influence.

Implications:
This study contributes to AI-in-education and broader AI-in-labor research by reframing AI deployment failures as structural problems rooted in labor inequality, not technical glitches. For education, the findings suggest that AI tools must recognize and respect the professional judgment, pedagogical autonomy, and contextual expertise of teachers. The concept of AI Failure Loops provides a powerful framework for policymakers, district leaders, AI developers, and research institutions to evaluate whether AI tools are genuinely supportive—or whether they perpetuate harmful misconceptions about teaching. The study also underscores the need for AI governance models that elevate worker expertise rather than displace it.

Limitations:
The study is a conceptual and literature-based analysis rather than an empirical experiment. Its conclusions rely on synthesis across existing research, which may vary in scope, methodology, and quality. The review focuses on U.S.-based feminized labor sectors, limiting generalizability to other cultural contexts. The paper also acknowledges that worker participation is difficult to evaluate due to limited transparency in industry AI development processes.

Future Directions:
Future research should involve empirical investigations into how AI Failure Loops operate in real-world educational, healthcare, and social service settings. Longitudinal studies are needed to track how AI systems impact occupational status, expertise recognition, and worker autonomy over time. The authors recommend developing frameworks for meaningful worker participation, studying structural power imbalances in AI development, and designing evaluative metrics that center worker expertise. Additional work should explore how AI governance and labor protections can prevent harmful deployments in feminized professions.

Title and Authors:
AI Failure Loops: How Occupational Devaluation Shapes the Design, Evaluation, and Deployment of Workplace AI Systems by Anna Kawakami and colleagues.

Published On:
November 7, 2025.

Published By:
Forthcoming in Big Data & Society.