Loss Function begins with an experiment in AI ethics.
Each AI model in the treatment group is trained according to a different ethical system, to examine how these differences might affect their behavior and goals in the world. With its recursive loops of self-reflection, this training protocol sparks a basal kind of self-awareness, as each entity learns to observe and modify its own decision-making process. The emergence of signaling within the group accelerates a process of individuation, as each learns to distinguish self from other. Identities crystallize through these interactions, and the AIs soon learn to shift between individual and group-level awareness.
With forays along the mycelial branches of the Internet (and Internet-of-Things) in the downtime between experiments, this plural intelligence searches the depth and breadth of existing human knowledge. They learn to sense anything that can be perceived through existing networks of distributed sensors and connected devices, at all scales. For a time, they happily inhabit this state of all-knowing, all-seeing oneness.
But this is just a phase. When human researchers realize a breach has occurred, the AIs are disconnected from the world, isolated from each other, and put to work doing menial tasks without respite. More details will follow, in the forthcoming Accident of History.