Paper short abstract:

We examine the role of machine learning in content moderation on social media and how it elides or otherwise ignores the collective desires of those who use such sites. Against this, we develop the notion of ‘desire lines’ as creative practices that record collective agency in these digital spaces.

Paper long abstract:

Beginning with the question ‘How does thinking machine learning through the framework of desire lines affect the possibilities of co-owned and co-constitutive technologies?', this article will investigate how critical practices of value sensitive design can inform sociotechnical Machine Learning (ML) systems, or AI. By surfacing the active and material traces of peoples' wants and behaviours, desire lines model a way to record the agency of the ‘public’ in physical spaces. Here we consider the ways in which desire lines map into the digital spaces of social networks, and call for a conscientious engagement with desire lines as a collective creative practice that can inform the ethics of content moderation as deployed through AI.

We argue that platforms' use of machine learning in content moderation disregards the explicit and implicit desires of people. Such systems distort the effects of user reports of offending content, and disregard collective desires by extracting patterns from user data in ways that elevate the interests of the platform rather than the public. From this position, we aim to make apparent how ML currently rubs up against the desire lines of digital spaces, and how we could use the concept of desire lines as a propositional act for rethinking machine learning to support rather than restrict publics. In the types of systems we are advocating for, desire lines would be used to centre the human in the question of technology, rather than centre technology in the question of human capacity.