Sociotechnical refers to something that is shaped by both social factors (like culture, politics, power structures, and human practices) and technical factors (like algorithms, machines, and infrastructure). It emphasizes that technologies don’t exist in a vacuum—they’re always designed, used, and understood within human contexts that influence how they work and whom they benefit (or harm).

Sociotechnical 

The article uses the example: Many mental health apps emphasise self-management approaches over contextual understanding in their design, deflecting attention away from structural injustices while presenting a veneer of technological empowerment.

Self-Management/Mental Health Apps Example

The authors argue that ableism in AI must be understood through an intersectional lens, drawing on insights from critical race, queer, and Black technoscience studies. Just as technology can reinforce racism, it also embeds and reproduces ableism—often in overlapping ways with other forms of oppression. This means tools developed to critique technological racism (and other oppressive systems) can also help uncover and challenge technological ableism.

Intersectional Roots of Technological Ableism

This phrase refers to the fact that many AI systems—especially machine learning models—are so complex that it’s very difficult to see how they reach their conclusions. Inputs go in (like data), and outputs come out (like predictions or classifications), but the decision-making process inside is often opaque, even to the people who built the system. This lack of transparency makes it hard to identify bias, explain errors, or hold designers accountable, and it can reinforce the illusion that AI is neutral or objective when in reality it reflects human assumptions and societal inequities.

"Black Box" Nature of AI

The authors critique the dominant view in AI fairness debates that bias is purely a technical problem, solvable through algorithmic “fixes” like debiasing methods. They argue instead that AI bias is a sociotechnical issue, deeply embedded in the social, cultural, and power structures in which AI systems are designed and deployed. Because these systems are created within inequitable contexts, they inevitably reinforce existing injustices, and their “objective” veneer masks the subjective, situated choices shaping them.

Arizona State University

The article argues that bias against disabled people in AI systems often begins long before algorithms are trained—it starts with how “disability” itself is defined. Drawing on disability models like the medical and social frameworks, the authors show that these definitions shape data collection, system goals, and deployment, often embedding ableist assumptions. They emphasize that true fairness requires transparency about these definitions, participatory design with disabled people, and a structural lens that moves beyond abstract metrics to confront cultural and systemic ableism.

Definition Drives Design: Disability Models and Mechanisms of Bias in AI Technologies 

Newman-Griffis, D., Rauchberg, J., Alharbi, A., Hickman, R., & Hochheiser, H. (2023). Definition drives design: Disability models and mechanisms of bias in AI technologies. First Monday, 28(9). https://doi.org/10.5210/fm.v28i9.12903

Definition Drives Design: Disability Models and Mechanisms of Bias in AI Technologies Reference

AI Bias as a Sociotechnical Issue

A central argument of the article is that how we define disability at the outset of AI design shapes every step that follows. Definitions guide what data are collected, how labels are assigned, and what outcomes the system is meant to optimize. These definitional choices are rarely transparent (lending themselves to what researchers call the "black box" nature of AI systems). Yet, they embed value judgments that determine whether AI technologies reproduce ableist assumptions or move toward more inclusive design.

Learn Before

Related

Learn After