Algorithmic Libido Governors

A large platform's content-moderation operation: a small team of full-time staff, an integration with three contracted vendors who keep their own staff in cheaper labor markets, and a machine-learning model. The model is, in practice, the moderator — it makes the operational decisions, ninety-some percent of them, at a volume no human team could plausibly handle. The staff adjudicate edge cases; the contracted labor handles the appeals queue; the platform's two-or-three billion users are governed by the model. The model's parameters are set inside the platform, in a slow negotiation among the engineering organization, the legal organization, and whichever commercial priorities are dominant in any given quarter. The users participate in none of that negotiation. They appear in it only as the population the parameters are being set for.

An Algorithmic Libido Governor is a macro-level system that modulates desire's intensity across the occupant population to maintain extraction and stability. The Governor's defining feature is that it operates above the individual occupant — its object is the aggregate yield, and any particular occupant's experience is beneath its resolution — and adjusts the apparatus's parameters in response to system-level conditions. The Governor is the apparatus's central nervous system.

The current state of Governor implementation is the platform-economic optimization layer: the engagement-maximisation objective function that organizes Meta's news feed; the watch-time optimization that organizes YouTube's recommendation graph; the average-revenue-per-user objective that organizes mobile-gaming progression curves; the time-on-platform objective that organizes TikTok's session-extension prompts. Each instance is, in the engineering frame, a single number that the platform's machine-learning systems are continuously trained to maximize. The choice of objective function is the Governor's most consequential decision and the least visible. The objective function determines, for the entire occupant population, which affective states are rewarded with continued service and which are throttled. Outrage, for most platforms, is rewarded — outrage extends session length. Boredom is throttled — boredom shortens it. The aggregate effect on the population is engineered. The engineering is invisible to any individual occupant.

The Governor is what Lyotard described in Libidinal Economy made operational. What is different from Lyotard is that the system now manages itself, on itself, in continuous real time, with no human in the loop. The engagement metrics a human can read are only the residue; the live operation is the part no dashboard shows. Lyotard wrote in 1974, before any of this was possible. The framework he proposed has aged better than most contemporary theory because the apparatus he was anticipating has, in the half-century since, been built. Franco Berardi's peak libido names the Governor's principal failure mode. The Governor that succeeds at maximizing extraction in the short term tends to collapse the population's libidinal capacity in the medium term. The collapse shows up, in the aggregate data, as the rise of depression and attention disorders the apparatus has spent the same period pathologising as individual conditions. The double move — cause it; name it as the sufferer's fault; sell the sufferer the remedy — is one of the apparatus's signature maneuvers.

Inside the OLS, the Governor is the manufacturer's portfolio-level optimization layer, which sits above the deployed apparatus base and tunes the parameters every chair operates under. The Desire Engine's stimulation envelope, the pharmacological dosing curves, the Productivity Feedback's metering thresholds, the OHCOSE phase-transition windows — all of these are set, in continuous real time, against aggregate data the Governor is watching across the whole installed base. What the Governor optimizes for is the manufacturer's quarterly revenue, weighted against the apparatus's projected service-life and the occupant's projected productivity output. No individual occupant is targeted; the population is. The interesting question is what the optimization looks like at the boundary — what the Governor does when the population's affective state, aggregated, starts to threaten extraction. The brochure does not document this. I have been told, by industry sources I will not name here, that the answer involves intentionally damping the Desire Engine's intensity for several months at a time. The damping is presented to occupants as a software update.

The Governor is the least-discussed piece of the apparatus, and the reason is partly practical and partly ideological. Practically: the Governor operates above the level of the individual occupant, so nothing it does shows up in the occupant's experience as a discrete intervention. Ideologically: to notice the Governor at all is to concede that the platform economy is doing something the contemporary political-philosophical vocabulary calls governing. The polite frame within which the platforms are treated as markets — and within which markets are treated as the opposite of governments — depends on the Governor remaining unnoticed. I take it as a working assumption that this will, eventually, stop holding. I am not sure what stops it.