Many have long worried that AI and robots will replace workers. But less attention has been paid to the increasing use of algorithmic systems to manage workers — creating ever more authoritarian and exploitative workplaces.
Workers fulfill orders at an Amazon fulfillment center on Prime Day in Melville, New York, on July 11, 2023. (Johnny Milano / Bloomberg via Getty Images)
From cab drivers and baristas to Amazon workers and McDonald’s cashiers, many workers are concerned about how new technologies are degrading labor conditions. Across the spectrum of politics and perspectives, many see automation and computerized decision-making as threats to the quality or existence of workers’ jobs.
But in his forthcoming book Cyberboss: The Rise of Algorithmic Management and the New Struggle for Control at Work, out from Verso in August, scholar Craig Gent argues that people often fail to understand the most harmful aspects of new technology in the workplace. He demonstrates that the overriding concern for most workers should not be replacement by robots but management by them.
Gent argues that “algorithmic management” — the use of computers to supervise worker productivity and make workplace decisions — worsens managerial structures that already rob workers of their agency and dignity. Resisting this heightened domination requires recognizing how algorithmic management undermines workplace organizing, and campaigning for its suppression.
Cal Turner and Sara Van Horn spoke with Gent for Jacobin about how algorithms are reshaping the workplace, the shortcomings of some union responses to AI so far, and the creative ways that workers are fighting back.
Sara Van Horn
Algorithmic management and algorithmic power are key concepts in Cyberboss. Could you define these terms?
Craig Gent
Algorithmic management essentially means that workers’ frontline experience of management is actually a computer, an algorithm — or, in more nebulous terms, a “system” — rather than a human manager.
Algorithmic power is a third plank of power within the workplace. In addition to managers and their power and workers and their power, the computer system itself has its own authority within the workplace.
Cal Turner
Could you give some examples of workplaces where technology is playing an increasingly managerial role?
Craig Gent
You see it everywhere, from fast food and cafés to taxi firms, newsrooms, and the justice system. But at the apex are logistical workplaces, where in addition to organizing physical processes of moving goods around, algorithms are managing the workers who work there. That can take the form of managing their movements, managing their tasks, or managing the allocation of labor itself.
Typically, it will look like workers are engaging with “the system” through a screen-based interface. For example, nowadays McDonald’s workers are interfacing with a screen that is organizing the orders in front of them. In logistics workplaces, it’s often a handheld scanner. Within the gig economy, it’s usually an app.
Sara Van Horn
Could you talk a little bit about the fallacies of the “robots are coming for my job” argument against automation? What does it misunderstand?
Craig Gent
People have been talking about workless factories — so-called “lights-out” factories, where you don’t even need to turn on the light because robots don’t need light to work — since at least the 1970s. Obviously we have seen well-publicized prototypes of robots running around workless factory floors. But robots are very expensive to maintain, fix, and procure, and they run into problems that humans don’t.
It’s much more likely that we’ll see workers managed by computers than replaced by them.
For example, the mundane problem of dust is a problem for robots that are spending hours a day roving around a warehouse floor. So I wager that we’ll see a truly workless warehouse around the same time that we master the dustless warehouse — which is not to say that it will never happen, but it’s improbable. It’s much more likely that we’ll see workers managed by computers than replaced by them.
Cal Turner
You write that there is more at stake in the future of work than remuneration and certainty, meaning that labor unrest often strains against not just low wages, but also managerial control. What does this mean for the future of work?
Craig Gent
It means that we will see a further assault on dignity at work and what it means to be respected as a worker. Few people really enjoy the fact that they have to go to work, but absolutely no one wants to be actively squeezed for their labor and treated in inhuman ways while they’re working. I think we’ll see a degradation of how we can expect to be treated at work, which has political implications beyond the workplace for class politics more broadly.
It also means that we have to think about labor organizing in broader ways than we’ve become used to. Many labor organizations like trade unions have over time shrunk what they’re about: they are about pay, pensions, terms and conditions, health, and safety. But in many cases, those concerns don’t really touch the essence of where and how algorithmic management enacts power.
Sara Van Horn
Could you talk more about that tension and how you have seen it playing out? How have unions responded to algorithmic management — or not responded?
Craig Gent
Speaking from a British perspective, unions are obviously aware that technologies are reshaping work, and they have suspicions that these technologies are being used to treat workers badly and to degrade the work itself. However, when it comes to the actual substance of their demands, they haven’t had much to say about the technology itself.
The reason for that is that trade unions over the last hundred years have gotten out of practice with laying any claim to the organization of work itself, and many don’t even see it as their role. The idea of the manager’s right to manage is strong even within trade unionism. Unions are happy to stake a claim on job losses or pay, but they see the actual organization of work as the business of managers. Landmark assurances of labor rights were gained on terms and conditions, but the explicit trade-off was jettisoning any claim over new technologies or reorganizing the work process.
The way that plays out is often that trade unions are forced to stand by and watch it happen, playing a buffer role and trying to improve the rate of exploitation. I remember speaking to one trade union official who was quite proud of the fact that their union had, along with the employer within a logistics center, ushered in independent time-and-motion studies to be carried out on the work. They saw it as an exercise in benchmarking: if time-and-motion studies could be done, then they would have a shared objective sense of what was possible within a given time.
That is completely buying into management logic, but they saw it as the union’s drawing a line in the sand around which they could negotiate. When I spoke to workers who had experienced such processes, it simply led to them having to work harder. Trade union officials understood time-and-motion studies in the workplace in a way that was completely antithetical to the benefit of workers on the shop floor.
The labor movement is also often uncomfortable with intervening because of the mediating role that they often adopt, with the workers on one side and the employers on the other and the union in the middle. Unions often want to temper forms of resistance and organization that aren’t the sanctioned ones. One thing that was really striking when I was speaking to workers was that they nearly all had an ambivalent relationship to trade unions — and they were doing the resistance anyway. If anything, they were resisting despite the trade unions that were recognized in their workplaces.
If we want to win on the terrain of algorithmically managed companies, if we want to find where the leverage is, then we’re not going to find it in the old ways of doing things. That is not to say that traditional forms of industrial action are pointless. But there’s this idea within a lot of unions that you must first organize and get recognition to negotiate, and if negotiation doesn’t work, then you have action short of a strike, and then you have a strike action, and then you get what you want. But that pipeline begins to break down in such a way that, if we’re really serious about getting ahead of this technology, we have to be far more creative in our approach.
Within very regimented ways of working, having a more tacit understanding of the ways that you can get around the system becomes crucial.
In studying how people are already resisting, I saw things like know-how and knack become really important. Within very regimented ways of working, having a more tacit understanding of the ways that you can get around the system becomes crucial. Despite the initial appearance that these are individualized forms of activity by disgruntled workers, there were forms of collectivity that sprang up around them, like workers sharing special codes to hack their wearable scanners.
We have to remember that managers are not at the top of the hierarchy. Usually, managers themselves don’t have great insight into the system. So if workers were able to exploit that fact — if tacit understanding and tactical know-how were combined with the institutional weight of a trade union movement that really wanted to shake things up — that would be genuinely impressive.
Cal Turner
Could you talk about AI specifically?
Craig Gent
AI means a lot of things in a lot of different contexts, but it basically means just very fast computers. Obviously, the AI that people became most familiar with over the last year is large language models, which in some instances — like we saw with the Hollywood writers’ strike — threatens to take away jobs. But it’s not really about management by computers. We could also call the systems that are used in Amazon fulfillment centers AI in the sense that they work cybernetically with a degree of autonomy.
AI has become relevant in conversations around algorithmic management because it’s something that trade unions are latching onto. In Britain, trade union efforts around AI are being led by the Trades Union Congress, which is the overarching body for most of the unions. They have a manifesto called “Work and the AI Revolution,” which is guiding a lot of the efforts of member unions around AI. But the manifesto really hinges on a thin ledge of health and safety and whether AI is going to — by virtue of its inhumanity — put workers’ health and safety at risk. It is also overly concerned with transparency in a way that misunderstands how algorithmic management becomes powerful in the first place.
The transparency argument comes from this idea that what’s problematic about algorithmic management is that it’s a black box: we can’t see how it’s making decisions, and therefore you can’t reason with it or make counterproposals. But if you open up algorithmic management, it’s just lines and lines and lines of code. There’s no portion of code you could hope to isolate that would account for the power relations in an algorithmically managed workplace.
Put another way: if you think the problem is a particular algorithm, and you’re not finding the social decisions that have gone into it, then you need to widen the picture. As the cultural anthropologist Nick Seaver writes, “Press on any algorithmic decision and you will find many human ones.” So the focus is essentially in the wrong place. The lines of code of an algorithm are not going to tell you what you want to know, which is really about the organization of work and power within the workplace.
Sara Van Horn
What would be the most strategic way for trade unions to engage with algorithmic management? What is the most powerful demand they could make?
Craig Gent
The most powerful demand would be suppression. I was given a lot of hope by the demands that the Writers Guild of America made around AI in its strike last year. They recognized that they couldn’t say AI was a health and safety issue, and instead argued that nothing good could come of AI within that particular sector.
I don’t think it’s possible to have ethical algorithmic management. Algorithmic management just intensifies labor — that is what it’s for. So suppression would be the demand that would make me really excited.
Short of that, however, I think serious attempts to undermine the power of companies that use algorithmic management would be useful, particularly because of the way that algorithmic management undermines things like strikes. I make an argument for broadening the historical organizing repertoire of trade unions, which is still based on a very twentieth-century version of the workplace. Instead, unions need to get more comfortable with tactics that aren’t new but that were used a lot more going further back in history, like subversion at work or sabotage.
The lines of code of an algorithm are not going to tell you what you think you want to know, which is really about the organization of work and power within the workplace.
One of the problems that unions have to overcome is the complete fragmentation of the workforce — not just in terms of precariousness, but within the workplace itself. Algorithmic management organizes workers within space such that they might not come into contact with each other, where work is a lonely and antisocial experience where you’re too strapped for time to get to know anyone. Algorithmic management robs us of the preconditions for organizing at work.
In terms of how algorithmic management undermines the strike itself, it’s as simple as companies being able to reroute around strikes. In Germany, during the first-ever strikes against Amazon, there were days and days lost to strike action. But the effect on Amazon was negligible, because it could simply reroute orders to other warehouses so fast that there was minimal service disruption.
Cal Turner
Could you talk about how people have resisted algorithmic management, both as individuals and collectively?
Craig Gent
There was a case of a slowdown at a warehouse near Heathrow Airport where workers were being tightly monitored for their productivity. Their shift allocation every day was based on whether they had achieved a certain productivity score the day before, so their day would start with a text message telling them whether their shift was confirmed or whether it was canceled.
These agency workers were paid 70 percent of what in-house workers were paid. A few of the agency workers got together and organized a slowdown where they decided they were going to work at 70 percent of the productivity target for the obvious symbolic reason. Because the workers were so used to having to work to a certain productivity metric that would be quantified and presented to them, they were actually able to gauge what 70 percent of productivity felt like. And this — their embodied relation to the productivity metric — was confirmed to them in the next day’s text messages.
Unfortunately, that action didn’t work, because the organizers got ratted out to management. But it was a really interesting moment in which people’s experience of working to the algorithm enabled them to resist it.
In other cases, resistance has taken the form of subterfuge, where people realize they can use the devices that they interact with day in and day out in unintended ways to get around the algorithm. They can take themselves out of productivity calculations, for example, or award themselves unsanctioned breaks, and then workers share this knowledge around on the down-low. In other places, resistance has taken the form of using computers and learned passwords to overcome the informational asymmetry that defines algorithmic management.