Software-layer behavioural modification has crossed the line from promising prototype to industrial process. Three case studies, one method.
For two decades the critique of "tech" was that it might one day shape behaviour at scale, and that this should be regulated before it happened. We are now well past the threshold the critique was meant to forestall. Engagement-ranked feeds restructured the news cycle and adolescent mental health. Algorithmic dispatch restructured how a substantial fraction of the working-age population gets paid. Compliance apps and risk scores have restructured policing, welfare, and immigration enforcement. None of this is speculative — the instruments work.
What this paper argues is narrower than the usual complaint. The usual complaint is that the systems are bad. The narrower argument is that the systems are effective, that the operators know they are effective, and that the public discourse has not caught up to what the operators have known internally for at least a decade.
Three domains — consumer platforms, gig labour, and state compliance systems — share neither owner nor ideology. They share a method: instrument the behaviour, score it, feed the score back to the subject in close to real time, and adjust the weights until the behaviour curve bends. The method is boring. The method is also, at this point, the most reliable lever any large institution has for changing what large numbers of people actually do.
Engagement-ranked feeds were sold as neutral surfacing of "what you want to see." They are a continuous, multi-armed bandit running on several billion people. The notable finding isn't that attention got broken — it's how cheaply, and with what precision, the breaking was achieved.
Rideshare, delivery, and warehouse work runs on a feedback loop the worker can feel but cannot see. Pay flexes by zone. Quotas adjust per shift. A rating slips and the next week of dispatch goes quiet. The most aggressive deployment of operant conditioning ever attempted on a non-prison population.
Welfare reassessment portals, risk-scored policing dashboards, immigration check-in apps, automated benefit clawbacks. The public sector spent the 2010s and 2020s adopting the conditioning techniques pioneered in the private sector — and now uses them on populations that, unlike app users, cannot uninstall.
What "the conditioning works" does and does not claim. Four obvious objections, answered up front. Editorial stance.