From back cover:
“The state will use machine learning to filter surveillance footage for criminals and potential criminals; capitalism will use machine learning to identify ways to appropriate resources and maximize profits…This doesn’t mean, though, that anarchists could appropriate machine learning systems for our own goals. Quite the contrary! For machine learning systems are not only at the beck and call of the forces of order through their emergence from prompts, but far more importantly, remain tethered to these forces through the models they create. The sources on which machine learning systems feed are the troughs of ‘big data’: billions of statistical, lexicographical, literary, medicinal, military and civilian, surveillance-based and contractually obligated, creative or robotic, data points.
…To sabotage a machine learning system, then, it must be fed content that is in itself coherent, very likely meets evaluation criteria, and yet leads the system’s propagations into a feedback frenzy from which they can’t escape. The machine learning system must be fed text that coherently self-destructs. And the text must directly concern the concept(s) that the machine learning system is modelling, to meet its evaluation criteria.”
