The emergence of a public space hosted on online platforms has seen the appearance of algorithms as mediators that filter, curate, and select the contents we see as users. As the entanglement between this digital public sphere and offline politics progresses, several works have addressed concerns regarding online segregation and polarization in social media. With a consensus settling on the existence of a global process of democratic decline connected with polarization and with the lack of trust in representation and institutions, fears regarding the role of social media in this process have also expanded. This article discusses a particular technical opportunity presented by the rise of AI systems as mediators. This opportunity leverages a parallelism between two theoretical and seldom connected perspectives: 1) spatial models of politics arising in political sciences, and 2) representation learning spaces ubiquitous in recent and growingly ubiquitous forms of AI. By laying a bridge between the two models, this article proposes that the emergence of AI also enables the possibility of delimiting and disentangling the part of computation related to politics (and potentially involved in political segregation and polarization), opening a path towards better tools for social platform and AI compliance, regulation, and design tools.