Limited Deterrence, Strong Preemption: Digital Surveillance and Collective Action in China

• In Event: How Repression and Resistance Take PLACE

Abstract

Does digital surveillance suppress protest in authoritarian regimes? Existing research produces inconsistent findings, in part because it treats "digital surveillance" as a single phenomenon, conflating technologies that operate through fundamentally different mechanisms.

We argue that surveillance enables two distinct logics of power. \textit{Disciplinary power} operates through visible monitoring infrastructures such as cameras: their salience induces self-restraint and micro-spatial adaptation among those who may be observed. \textit{Identification-and-intervention power} operates through algorithmic tools such as facial recognition: by converting behavioral traces into person-level identifiability, these systems enable authorities to preemptively target organizers before mobilization scales. The two logics are distinct but complementary, as cameras generate the data that algorithmic systems require.

We test these arguments in China by applying computer vision to 8.2 million street-view images at 577,637 GPS locations to construct a micro-geographic camera dataset, which we link to 136,330 geocoded protest events and city-level facial-recognition procurement indicators.

Facial-recognition capacity is strongly associated with fewer protests at the city level, and its effect is largest where camera infrastructure is densest, consistent with complementarity between data capture and algorithmic identification. Camera density alone has no reliable aggregate effect. GPS-level evidence explains why: cameras suppress protest in the most intensely monitored cells, but the effect is confined to the 0--200m band, with suggestive short-range displacement at 200--400m and no effect beyond. Cameras reshape protest geography at the block level without reducing city-level counts.

The findings point to an integrated surveillance stack rather than a single deterrence channel. Disciplinary power operates at the micro-spatial scale; algorithmic identification produces the larger aggregate protest declines. The question is not whether surveillance deters, but how cameras and algorithms combine to produce discipline and preemption at different spatial scales.

Author

• Han Zhang, Brown University