Algorithms and worker power
In AI circles, a “centaur” describes a certain kind of machine/human collaboration, in which “decision-support” systems (which the field loves to call “AI”s) are paired with human beings for results that draw upon the strengths of each, such as when a human chess master and a chess-playing computer program collaborate to smash their competition.
In labor circles, “chickenization” refers to exploitative working arrangements that resemble the plight of the American poultry farmer. The U.S. poultry industry has been taken over by three monopolistic packers, who have divided the nation up into exclusive territories, so that each chicken farmer has only one buyer for their birds.
Farmers are “independent small businesspeople” who nominally run their own operations, but because all their products must be sold through a single poultry processor, that processor is able to exercise enormous control over the operation. The processor tells the farmer which birds to raise, as well as what the birds are to be fed, how much, and on what schedule. The processor tells the farmer how to build their coops and when the lights are to go on and off. The processor tells the farmer which vets to use, and tells the vets which medicines to prescribe.
The processor tells the farmer everything…except how much they’ll be able to sell their birds for. That is determined unilaterally when the farmer brings their birds to market, and the payout is titrated to the cent, to represent exactly enough money for the farmer to buy birds and feed and vet services through the processor’s preferred suppliers, and to service the debts on the coops and light and land, but not one penny more.
Chickenization has spread beyond chicken farming. Uber drivers are paid on a variable reinforcement drip-feed that gives them just enough to keep up the lease and gas and insurance payments on their vehicles, but not enough to give them breathing space to think about changing careers.
Likewise for Amazon drivers. Amazon styles the drivers as subcontractors working for a “Delivery Service Partner” (DSP), and the DSP is fully chickenized: They have to buy Amazon vans and subject them to Amazon maintenance, but Amazon reserves the right to fire a DSP without notice, stranding them with vehicle and lot leases and payroll liabilities.
The DSP owners are chickenized, but the drivers themselves? They’re more like the chickens. Or rather, they’re centaurs: From the instant you get behind the wheel of an Amazon van, you are being surveilled by an array of cameras hooked up to high-handed, judgemental AIs that monitor your facial expressions, your eye movements, and your ability to meet an impossible quota.
But even though an Amazon driver represents the tight coupling of a human and a machine to do more than either could do on their own, that’s not the kind of centaur that we talk about when it’s a chess master paired with a chess program. That chess master is being augmented by the machine, and the machine is the junior partner in the relationship. The human is the head, and the AI is the body.
By contrast, an Amazon driver is a reverse-centaur. The AI is in charge, and the human is the junior partner. The AI is the head, telling the body what to do. The driver is the body — the slow-witted, ambulatory meat that is puppeteered by the AI master.
The next generation of labor exploitation merges chickenization with reverse-centaurs. DoorDash and other gig companies use apps to script the movements and conduct of “independent contractors” to the finest degree, while hiding their true wages from them until they’ve finished their jobs.
The simplistic answer to this is, “Well, if you don’t like it, don’t work for a gig company.”
But that’s not how labor rights work. Behind every labor struggle is a recognition that business owners have negotiating power over workers that will, over time, shift more and more value from workers to bosses. Labor market outcomes are not a matter of individuals bargaining with one another: In a labor market, atomized workers bargain with consolidated firms.
It’s hard for workers to play companies off against each other to get higher wages (and companies get furious when they pull it off), but it’s easy for businesses to play workers off against each other to pay lower wages.
Historically, worker power has come from unions, which were able to bargain for workers at a specific job-site, or across a sector, and, just as importantly, to win labor regulations that protect all workers. Forty years of all-out assaults on labor organizing have weakened that source of worker power, and both Republicans and Democrats are deeply committed to keeping unions weak.
Despite this lack of institutional support, unions are coming back with a vengeance, and they’re pulling it off by creating new, high-tech organizing tools and strategies that answer high-tech employer tactics with high-tech worker tactics.
Using messaging tools to organize unions is an effective but indirect answer to chickenization and reverse-centaur tactics. A new kind of tech-based labor tool confronts these tactics head-on: counter-apps that help workers seize the means of computation from their bosses.
Take Para, a collection of apps for gig workers that undo the reverse-centaur relationship between workers and gig employers. For example, when DoorDash sends a job offer to a “Dasher,” Para unwraps that offer and reveals the total compensation on offer, which the DoorDash app hides by default until the job is done.
DoorDash understands that its customers want cheap deliveries, and it understands that its drivers don’t want to service cheap jobs where they actually lose money after factoring in time, gas, and depreciation. By hiding the total compensation from drivers, DoorDash is able to trick some of them into making unprofitable runs, allowing the company to offer customer-attracting loss-leader prices that are billed to drivers.
That’s where Para steps in. Para’s long-term goal is to build apps that automate the process by which workers play gig companies off against one another. That way, the auction process, where workers bid to see who will do the job for the lowest pay, is replaced by a process where employers bid to see who will pay the most for a worker’s time.
That’s a powerful vision, but it’s only part of the answer to the chickenized reverse-centaur industry. After all, there’s only so much bargaining power an individual worker can exert. Groups of workers, on the other hand, have a long track record of wringing concessions from employers.
Dashers understand this. The #DECLINENOW movement organizes Dashers (via online forums) to turn down all jobs below a certain payment threshold, with the understanding that the DoorDash algorithm will bid up the offered payments until a worker agrees to do the job.
That’s a process that’s ripe for automation — imagine if Dashers could form co-ops around apps that hid job offers unless they came with a profitable payout.
Counter-algorithmic activism is subtly different from Para’s model: While Para exposes hidden information and automates the process of uncovering it, counter-algorithmic work actually seeks to turn workers into centaurs, not reverse-centaurs.
For example, in 2020, there was a widely shared story about Amazon drivers hanging their burner phones from tree branches near Amazon warehouses. They were doing this because the Amazon delivery algorithm required them to make impossible quotas — impossible in part because of the number of deliveries they had to make, but also impossible because the system wouldn’t allocate a delivery to them unless they were close to the warehouse. By hanging their phones from trees at the warehouse gates, the drivers were able to trick the system into allocating deliveries to them.
This is a crude hack, but in Indonesia, it’s been elevated to a science, thanks to the proliferation of tuyul apps. These are apps created by and for delivery drivers that modify the official delivery dispatch apps. One key tuyul feature: spoofing the GPS telemetry fed to the dispatch algorithms so drivers don’t have to enter dangerous traffic jams around train stations in order to book rides with commuters as they arrive.
In other words, tuyul apps dispense with the fuggly hack of hanging phones from trees, and replace it with transforming phones to act on behalf of workers, not their bosses. (Tuyul apps do a lot more! Check out Rida Qadri’s fantastic reporting on the subject.)
Labor movements have done the most good when they build solidarity across sectors and industries. Lucky (?) for the chickenized reverse-centaurs, there’s plenty of people finding themselves in that circumstance.
Take Taylor Lorenz’s excellent Washington Post article on “algospeak,” an emerging dialect of euphemisms that social media users deploy to bypass the filters that the dominant platforms use to keep discourse brand-safe and minimize publicity scandals.
Much of that article dealt with the problems all communicators face. For example, if you want to talk about suicide and mental health, you have to use the euphemism “become unalive” because the filters downrank any mention of suicide.
But algospeak isn’t just a communications issue: It’s a labor issue. The people who truly live and die by algorithmic ranking choices are the people whose ability to put groceries on the table is directly tied to whether a social media platform suppresses their videos or text.
These video creators are also the source of the bulk of the platforms’ wealth, but — like other chickenized workers — they have no way to know what they’ll get paid for their labor, and don’t find out until the algorithm digests their materials and decides whether or not they will be exposed to the people who’ve explicitly subscribed to their feeds.
These workers are also allies in the fight to create true centaurs out of reverse centaurs. That’s where Tracking Exposed comes in: They’re a loose collective of digital human rights advocates who offer browser plugins that help reverse-engineer the recommendation systems of Facebook, Pornhub, Amazon, TikTok, and YouTube.
These are producing data that creators can use to “walk without rhythm so they don’t attract the sandworm” (AKA the algorithm’s downranking fist). This is fully compatible with the Online Creators’ Association’s demand for “Transparent and Responsive Moderation.”
But these analyses are also yielding new recommendation systems: Tracking Exposed’s YouChoose tool replaces YouTube’s recommendations with recommendations from across the web, which you can understand and modify.
It’s not just online creators and Amazon drivers and Dashers who have natural solidarity when it comes to rendering algorithmic judgments legible and accountable. Tracking Exposed’s other work includes detailed analysis of the way that these systems influence elections and how recommenders create “an alternate universe” in autocratic, warlike states.
The pandemic was a boon to the bossware industry, as locked-down employees discovered how quickly “work from home” could turn into “live at work.” The promise of bossware — tools that monitor your face, your fingers, your conversations and more — is that someday, all workers can be chickenized reverse-centaurs.
The project of understanding and then seizing the algorithm couldn’t be more important to these workers, and soon enough, to all of us.