Meta-influencers and automation as wedge

Meta’s decidedly non-buoyant test balloons for a new class of AI-driven fake accounts, rediscovered this week in a bevy of profoundly cringe-worthy Instagram screenshots, don’t make much sense at all in terms of what platform users might ever want to experience. They make a good deal more sense from a vantage point of labor flows and historical automation trends. Based on those patterns, we should expect that Meta will sooner or later invite influencers—the suppliers in the major pipeline of creative work that keeps audiences on its platforms—to accept in-house roles as meta-influencers (as it were) whose work will consist of adjusting the parameters for various AI-driven accounts.

Meta depends on many types of labor, but influencer labor is probably the type where they see the most liability that someone outside Meta might exert control over the labor arrangement. That is, influencers have a much greater (hypothetical) capacity to withdraw their work from Meta’s platforms in favor of rival services than do content moderation workers, whose labor is sold in bulk to a platform company of their employer’s choice. While I’m not aware of any movement toward collective lobbying on the part of Instagram or Facebook influencers so far, Meta presumably has a close eye on this dependency and on examples from other platforms (Spotify and Substack come to mind) where prominent suppliers have withdrawn their work to protest company policies.

This type of situation—when a skilled worker class might exert real leverage over a company if they got around to organizing—is where industrialists historically make automation a priority. Contrary to the main tropes espoused by boosters and critics alike, automation (including AI) almost never replaces workers. (Hence why its effects on employment and productivity don’t really register at an aggregate, political economic level, as Aaron Benanav has shown.) Automation is better understood as it was characterized at its factory floor origin by Marxist critics like James Boggs and Charles Denby: as a multi-pronged form of sabotage-from-above to shake up an arrangement where workers held too much power and where new machinery offered a pretext to relocate, reclassify, disorient, and otherwise break solidarity among workers. Historians of technology including David F. Noble and Wendy Hui Kyong Chun have since captured automation and computerization’s more subtle intervention, the one that always attends the label of programming (whether parts programming for 1950s machinists in Noble or computer programming in Chun): it splits works processes into intellectual and physical steps, abstracts the intellectual routines, and empowers the select workers who now control the latter while also distancing and alienating them from their old work context. The figure I’ve used with students in my automation/AI course is that automation is a wedge: it intervenes in a working context by pushing a select few workers upwards and deskilling or precaritizing the rest.

After the public backlash to Meta’s AI accounts, it’s not inevitable that they will roll out anytime soon, but if and when they do, here’s my bet as to how it will go: Meta won’t try to replace influencers but will instead bring a select few of them in house as meta-influencers. The pitch is that instead of precariously managing your own, individual identity as an influencer, you put your creative savvy to work in managing (by writing text prompts) a set of identities for automated influencers. The appropriation and cringe factors are considerable for us seeing this week’s screenshots, but they mask some appeals that influencers might take pretty seriously: you move from an extremely precarious self-employed position to a salaried gig with great benefits; you no longer need to sacrifice your own expectation of privacy in order to pursue “influencing” as your calling; and you get to flex your creative skill within that calling by applying it to a wider variety of characters beyond just yourself. These AI accounts wouldn’t be very “organically” popular, in platform industry terms, but Meta would readily inflate their visibility in user feeds while hiding posts from old-fashioned influencers (offering user safety as a rationale if they bother to offer one).

This pattern is essentially how automation took hold in music radio in the 1960s and 70s. Stations didn’t replace DJs with robotic counterparts overnight; the machines were still too expensive and unsophisticated to replace anyone, requiring at least as much labor to maintain and program them as they might have reproduced. Instead, station managers and automation designers pitched DJs on the superior—notably, more “creative”—work of musical programming.¹ Live broadcasting, according to radio automation’s boosters, was tedious and understimulating, subordinating DJs to the record player and making them wait idly as songs played; by interfacing instead with an automation system that would play back records in the sequence they programmed, they could set their sights on the larger countours of the station’s sound. To station management attending the National Association of Broadcasters conference, the actual rationale was made plain: by distancing DJs from the live transmission chain, automation could minimize the pervasive risk that DJs might deviate from management’s dictates about music selection and advertiser-friendly decorum. A pre-programmed radio show was one that managers could review ahead of time, even if the person doing the programming really did experience it as more creative. And a considerable number of DJs did accept the offer to become broadcast programmers, even as others—like April Feld—saw through the cynical framing and penned obituarites for a medium that had rejected them.

My point in analogizing radio automation to AI-driven Instagram accounts is that the same basic elements are in place: a stabilized media industry, a (potentially) unruly creative worker class providing its value, and a technological novelty. It doesn’t matter that the novel technology offers zero appeal to end users, nor (contrary to most of the reporting on this) do the tech’s actual capabilities and limitations particularly matter. Radio audiences of the 1960s were no more eager to hear clunky automated song transitions and lose their line of communication back to stations than Instagram users (credulous tech reporters aside) are to engage fake influencers in meaning-free DM chats. But platform firms like Meta have more than enough power to impose something like AI accounts on audiences, and influencers should get ready sooner rather than later to face a more concerted rollout of this feature—a rollout that will offer some of them a new kind of job and steadily siphon away livelihoods from the rest.