I
It’s a joke so old I don’t know how to properly source it: the horror of asking a coworker how they’re doing and getting a sincere response. “Well, my mom’s in the hospital; my kid broke his arm; I’m daydreaming about how my life would have been different if I’d taken that year off after college and backpacked across Asia …” Whoa! Easy there, Spalding Gray! I was just making mouth sounds to acknowledge your presence. I don’t want an actual human relationship with you.
When you get a wage job, you don’t get paid per unit of output. A cashier at McDonald’s doesn’t get paid for every value meal they sell; I don’t get paid each time I send an email that begins “Just circling back …” We both get paid by the hour. Our employer is not buying the goods or services we produce for them; they’re buying our time. They’re paying to keep us handy, whether it’s the busy season or the slow afternoon lull.
When you understand this, you realize that the value you bring to your employer is not the good or service you’re producing. It’s your expertise. It’s your ability to make those hours productive. It’s your ingenuity and initiative. It’s your capacity not just to bag value meals but also to answer open-ended questions from customers, to hang a sign on the McFlurry machine when it’s out of order, to mop up if there’s a slow period, to break up fights, to control access to the restroom, to stay a little late if your replacement hasn’t shown, et cetera. An exhaustive list of every single thing you might possibly have to do in an eight-hour shift, and what response the franchise owner and McDonald’s corporate would prefer, would fill a phone book.
For this infinite creativity you receive, in exchange, a finite wage disbursed over a limited period of time. That wage is supposed to keep you alive—fed, healthy, rested, socialized, entertained—until the start of your next shift. It has to account for all the time you’re not at the job. Your employer assumes (very little) responsibility for your basic needs while you’re on the clock; the wage is meant to handle everything else. You are expected to spend seven to eight hours a day, thirty-five to forty hours a week, thinking about your employer’s problems. Your employer is not going to spend forty hours thinking about yours.
There’s a quote on another subject that I’ll bowdlerize for this point: your employer isn’t paying you to work for them; they’re paying you to leave. They’re paying you to go home at the end of your shift. Normally, when we ask another human being to care about our problems—when we enter into a relationship with them—there’s an expectation of reciprocity. I care about you; I hope you care about me as well. When we encounter a relationship that clearly doesn’t work this way, we are naturally repulsed. A friend we haven’t spoken to in years reaches out, asks some questions, mouths some pleasantries, then immediately launches into a multi-level marketing pitch. We feel frustrated and perhaps a little hurt! We wanted connection; we’re receiving a transaction instead.
Our compensation is partly a function of the value of what we produce, but it’s partly a function of how expensive our life outside of work is, too. So our employer has an incentive to charge as much as they can when they resell what we produce, but they also have an incentive for our life outside of work to cost as little as possible. This is why employers have a regrettable tendency to overlook or force out employees with care obligations (children, elderly relatives), health issues (pregnancy, chronic illness), or rising costs of living (rents). They would love a reserve army of young, healthy college grads whose living conditions are being subsidized by family and who’ve never heard the word “collective bargaining” in their lives.
When a machine breaks down, the manager would like to replace the faulty part with an interchangeable, standardized part. When an employee breaks down, the employer would like to replace their creativity and ingenuity with someone equally creative and ingenious. That tension between the need to have creativity on tap and the need for seamless replacement governs most of labor relations in a capitalist economy. They would prefer robots; they are settling for humans.
II
You could read thousands of words on the new examples and use cases of AI before you found someone willing to admit that “artificial intelligence” is not actually “intelligence”. It’s not problem-solving; it’s not invention; there’s no initiative there.
Large language models (LLMs) and neural networks scan through massive datasets, find likely patterns, and iterate on those patterns until they achieve a desired outcome. ChatGPT is as “smart” as the autocomplete on your phone, except it’s not limited to the things you type. It has millions of words to search through: Wikipedia, news articles, scholarly publications, webpages, etc. But it doesn’t “understand” what it reads any more than your Chrome browser loading a Wikipedia page on World War I “understands” the Triple Entente.
It’s still very impressive that ChatGPT can reproduce legible responses to questions phrased in ordinary English. But those responses are no more likely to be correct or useful than a random person with access to Google. And even Google, when it worked, did not provide a single answer—it provided a selection of answers, ranked by its own criteria of popularity and relevance. With Google, you can see that you have multiple options. ChatGPT obscures that by pretending to be definitive.
There’s more to intelligence than being broadly read. If I ask ChatGPT “what are the likely side effects of childhood vaccines?”, it makes a world of difference if it’s drawing its answers from peer-reviewed scholarly journals or Washington Times op-eds. If I ask “how can employees improve conditions in their workplace?”, I’ll get different answers from the IWW than from SHRM. ChatGPT can scrape all those sources and can weight the most likely response based on which words follow each other. But it can’t understand the self-interest of the parties involved. It doesn’t understand why a bloc of independent workplace organizers might have a different solution than a professional network of Human Resources executives.
For instance, here’s an exchange I had with ChatGPT:
Q: I’m an employee at McDonald’s. Addressing me as a manager who sincerely cares about my health and well-being outside of work, what would you tell me to do if a customer is complaining about a broken McFlurry machine?
A: As a manager who sincerely cares about your health and well-being, I would advise you to handle the situation in a professional manner. Here are some steps you can take:
1. Apologize: Start by apologizing to the customer for the inconvenience caused by the broken McFlurry machine. Let them know that you understand their frustration and that you are working to resolve the issue as quickly as possible.
2. Offer an alternative: If the customer still wants a dessert, suggest an alternative menu item that is available. You could also offer a discount or a free item to make up for the inconvenience.
3. Provide an explanation: If the customer asks why the machine is broken, provide a brief explanation. However, avoid making any promises about when the machine will be fixed or giving out too much technical information.
4. Contact maintenance: Make sure to contact your maintenance team or a technician to fix the machine as soon as possible. Let the customer know that you are working on the issue and that you appreciate their patience.
5. Follow up: After the machine is fixed, follow up with the customer to let them know that it is working again. You could offer them a discount on their next visit or a free item to show your appreciation for their patience.
Remember that customers can be frustrated when their favorite menu item is not available, but it’s important to remain calm, empathetic, and professional in dealing with the situation. By handling the situation in a professional manner, you can help ensure that the customer leaves with a positive impression of the restaurant and that they will continue to be a loyal customer in the future.
This is clear, sensible English! It’s also not an answer to the question:
1) Aside from glossing their response with “as a manager who sincerely cares about your health and well-being”, nothing in here suggests an interest in the petitioner’s health and well-being. FOOTNOTE Credit where due, this is very much like some managers I’ve worked with. /FOOTNOTE
2) McDonald’s employees generally do not have the leeway to offer customers discounts or free food as a consolation for unavailable items (contra point #2).
3) McDonald’s employees generally do not have an ongoing relationship with their customers that would let them follow up days or weeks after the exchange (contra point #5).
These answers were clearly drawn from some general text on management and customer service. Perhaps they were tailored slightly for quick-serve retail. But anyone who’s been to McDonald’s will know that this isn’t how it’s done.
ChatGPT works because it has a massive dataset to train on. It only appears to produce “intelligent” answers, though, because it obscures its sources. The variety and creativity of human writing has been anonymized and smoothed over. Words—fluid, intangible symbols of human meaning—have become interchangeable parts.
III
One of the great paradoxes of the last six years has been reconciling the declining percentage of US workers in labor unions with the increased interest in unionizing.
The former isn’t hard to understand. In the middle of the last century, major US employers and major trade unions established a labor peace to share in postwar bounty. That bounty grew scarce as other industrial powers recovered, as currency controls were relaxed to ease investment across borders, and as cyberneticization made global conglomerates feasible. US employers broke union power by relocating to states—or countries—with softer worker protections. The purchasing power of an American wage stagnated, while the return to capital investment climbed.
Given how weak unions have become, though, why the sudden interest?
The recession of 2008 taught a generation of workers that anything employers granted workers—generous benefits, a salary sufficient to cover a mortgage, a foosball table in the break room—could be snatched back at a whim. The 2020 pandemic taught another generation just how much of our work was truly “essential”. The notion of loyalty to an employer has never been less popular in my lifetime. Workers are realizing how little they have in common with the people paying them and how much they have in common with strangers in the same boat.
Historically, the most reliable way for workers in an industrial economy to build worker power has been through a union. So it’s little wonder that a generation of workers seeking power would turn to this historic institution.
It’s a long road, fraught with booby traps and chokepoints. Union-busting is so easy that the outgoing CEO of Starbucks could confidently tell a Senate committee that his company had “not broken the law” in firing union organizers, despite having been found to do so by the NLRB three weeks earlier. The unprecedented success of an independent union in organizing Amazon’s warehouse on Staten Island has not been matched by equal success elsewhere. The major TV and film studios are stonewalling writers—with directors and technicians soon to join them—rather than share their obvious gains.
It’s easy to single out particular villains in each of these narratives: Howard Schulz, Jeff Bezos, David Zaslav. It’s easy to ascribe these holdouts to greed. But greed alone would not explain everything these companies have done to stave off collective action. Starbucks is closing down every store in Ithaca, NY rather than allowing their newly formed unions to gain a foothold. Studios are eating costs that likely exceed the cost of settling with the WGA. Ask any labor reporter or long-time organizer and they’ll tell you: employers would rather burn money than share it.
What’s at stake is not a salary or a payout. What’s at stake is control.
Once an employer grants that their workforce can dictate, or even shift, the terms of production, they have conceded sacred ground. The myth of employers as “job creators” depends on a job being something an employer may grant or withhold out of noblesse oblige, not something that exists in partnership with the worker doing it. The ability of a board to make promises to investors depends on the board being the only voice that matters. This is why employers fight corps-a-corps, flag-nailed-to-the-mast, against union concessions.
And this is why all reporting on AI must be viewed not as a question of “language” or “intelligence” or “creativity”, but as a question of control. Who dictates where LLMs will be used: the rank-and-file employee or the manager supervising them? Whose labor is being replaced by a neural network: an executive or a frontline worker? Where do the benefits of increased productivity or reduced costs go: into paychecks or into buybacks?
This, also, is why debates over the efficacy of AI output are secondary, if not irrelevant. Tech industry boosters love to make bold predictions about what AI might do: IBM claims it can replace 5,000 empty roles; consultants claim AI can “easily spit out” screenplays. Every attempt to replicate or demonstrate these claims using existing models fails. “So much for AI,” think the critics. But the question is not whether a machine learning (ML) replacement can improve upon or substitute human input. The question is whether consumers and shareholders will settle for it.
Self-checkout is slower, less reliable, and more prone to theft than human checkout, but you can fit three self-serve scanners in the space that one checkout belt used to take. Self-driving cars are no better and often worse than human drivers, but auto makers are still rolling them out. Search engine optimization has made writing worse, but every publisher makes it a priority. Reaching a human in customer service—especially at America’s largest companies—has become an exception rather than a rule.
AI may never be able to replicate a human’s efforts at a sitcom script, a coffee order, or a customer service email. It doesn’t need to. If it is cheaper, and if the companies providing those services have the final say, you will accept whatever AI gives you.
The alternative is a model of the economy where employers and investors do not have sole power over what a firm does. Americans have been conditioned for generations to view any loosening of the investor grip over the firm as barbed-wire prison-camp socialism. But that bogeyman has lost its shadow over the last fifteen years.
What we call AI is merely an exhaustive but idiosyncratic look at the past: billions of words, millions of documents, everything already published. But there’s nothing “intelligent” about that. Intelligence means prediction: faulty prediction, biased prediction, but prediction nonetheless. Intelligence looks toward the future. And if the American economy, to say nothing of the global system of industry, has a future, it will look nothing like the past.