Triggering Discovery with Anomaly Detection at the Large Hadron Collider

At the Large Hadron Collider, the CMS experiment selects events in real time via its trigger system, keeping only those that match signatures conceived by theoretical expectations. The rest are discarded; thus the trigger biases discovery towards physics it was designed~to~find. We~trace how discovery has narrowed to this rigid conception across three episodes: the~positron discovery, the establishment of Standard Model in the 70s, and the discovery of the Higgs boson. Each episode is read through the discovery philosophies of Feyerabend, Kuhn, and Lakatos. We show how from the 1970s onwards, the concept of discovery grows theory-laden and the method becomes rigid, until it institutionalises itself in how current experiments filter their data. We argue that anomaly detection through machine learning reverses this narrowing by fixing a measure of what is typical physics and letting the data decide what departs from it.