Travelling ML Models Across Different Socio-Technical Contexts and Purposes

The design and deployment of artificial intelligence (AI) often rest on specific cultural assumptions, perpetuating one-size-fits-all approaches and systems of oppression when applied uncritically across contexts (Birhane, 2020; Gabriel, 2020; Milan & Treré, 2024; Gammelgaard et al., 2023). While AI aspires to universal generalization, its outputs depend on data sources and design choices rooted in specific socio-cultural, economic, and technical settings (Chun, 2021). These values and assumptions shape how models function and make decisions. When transferred across geographies or repurposed for new tasks without translation, AI risks reinforcing inequities, erasing local specificities, and embedding hegemonic norms. This paper examines how cultural values, positionality, and problem framing influence technological systems, stressing the dangers of models that travel without translation (Lu & Qiu, 2023). Building on science and technology studies (STS) and the philosophy of computer science, we extend the concept of travelling technologies (Rottemburg, 2002), proposing a framework for evaluating models through the dual lenses of context, the dynamic sociotechnical configuration of norms, values, institutions, regulation, and infrastructures (Suchman, 2007) and purpose, or the aims of a system and the project behind it. Through these axes, we develop four scenarios of technological travel, linked to the technical elements implied in redesign and reconfiguration (Facchini & Termine, 2022; Buda, Manganini, & Primiero, 2025). The framework helps trace epistemic and material consequences of technological displacement, contributing to debates on coloniality and technoscience, and advocating for more just, situated approaches to AI.