The Care-full Stack: Collective Machine Teaching for Convivial Infrastructures

Algorithmic infrastructures can support fair, convivial and life-affirming ways of governing shared urban resources. However, more often than not, they are reshaping bodies and relations to accommodate extractive AI economics, optimise away joy and amplify injustice. The difference lies in who builds and maintains these infrastructures, and their underlying political and economic motivations.

Commoning is increasingly recognised as a viable socio-economic paradigm rooted in conviviality and care. In commoning, the community which benefits from a shared resource is the same community that stewards the resource’s production and management; it has a direct stake in regenerating it. The community has autonomy over the processes, protocols, and technologies of commoning and gears all layers of the commoning infrastructure toward retaining value within the commons rather than extracting value. This makes commoning a fruitful setting for imagining and rehearsing non-extractive algorithmic infrastructures. Approaches to care-based algorithmic infrastructures need to ask how predictive, generative or agentic AI layers inherit care from other layers of the infrastructural stack, rather than undermining it.

Viktor's presentation outlines collective machine teaching, an approach that enables communities to translate the convivial and affective aspects of community life into algorithmic layers of the infrastructure. The presentation introduces 7+1 steps of collective machine teaching, drawn from Scaling Material Urban Commons, a research project that won an award at the 2025 Media Architecture Biennale. It then reflects on the data-labour-related costs of collective machine teaching, the limits of scaling, and the role of procedural transparency.