Bosses and bots

How unions are dealing with AI in labour

Justin Gniposky first encountered artificial intelligence (AI) sometime between 2021 and 2022. He read an article about the potential and, like millions of other people, he started feeding prompts to ChatGPT, seeing what kind of stories it could come up with, and what kind of information it could sort.

His initial experience with AI tools didn’t raise any huge alarm bells, but that soon changed. As the national director of organizing for Unifor, he’s in touch with colleagues across many industries and organizations and leads a team of organizers. Very quickly, he started hearing people voice their concerns.

“I realized it was going to take a real grip on the issues we’re facing,” he says. People started asking, “How is this being used to further alienate workers and create a workplace that breaks down solidarity?”

Large corporations like Amazon leverage algorithms to disseminate anti-union propaganda to their workforce.

Employer-enforced alienation is a tale as old as class-stratified society. The nature of the capitalist mode of production has driven innovation and subsequent alienation in tech as much as it has in any other industry. In the late 19th century, Frederick Taylor devised a system of scientific management that he introduced into manufacturing industries to enhance labour productivity. Taylorism broke down the production process and kept each worker focused on a discrete task within it. Technological innovation has further refined this process through automation while allowing managers to surveil workers and enforce discipline on the job.

The tradition of alienating workers has carried forward into today, including via modern technologies. Large corporations like Amazon leverage algorithms to disseminate anti-union propaganda to their workforce. With generative AI, companies are able to constantly produce new workplace propaganda and subject workers to it.

According to Gniposky, anti-union messaging is ubiquitous within these workplaces, particularly in spaces where public messages are shared. These messages are supposed to contain important notices passed along by team members, but he notes that from what he’s heard, workers don’t see any positive union messaging posted there. For Gniposky, this begs important questions: “Who’s selecting the messages? Who’s drafting them? It’s clearly part of a system.”

Job Displacement

The fear of job displacement is one of the biggest buzzes currently around AI. In May, IBM cut hundreds of HR jobs in which the tasks could be easily automated. These layoffs take place amid a larger cutting back of the tech industry this year: Blue Origin laid off 10 per cent of its workforce, Bumble 30 per cent, Business Insider 21 per cent, and Dropbox 20 per cent, while Google offered early retirement to its Search, Ads, and Commerce teams. With tech workers already seeing the consequences of precarity in their industry, it’s unsurprising that AI anxiety is affecting these workplaces.

These losses, scalebacks, and anxieties are mostly in tech, particularly around human resources and administration.

Christopher Leinonen, an animator and member of International Alliance of Theatrical Stage Employees (IATSE) 938, has worked for two animation studios that have ratified union contracts, both of which included language around the use of generative AI for the first time. Even so, full replacement isn’t as near as it seems.

While many people and publications predict the demise of creative industries, Leinonen has yet to see an immediate major impact on the industry outside of generalized anxiety. “The studios that I’ve worked with here haven’t adopted it yet in any real capacity and the stated policy has been that it isn’t being done.”

At least one of the two companies pushed back on the AI language proposed during contract negotiations. “There was hesitancy on the part of the studio to include it in the agreement. That was something the bargaining committee had to fight for,” Leinonen says.

While generative AI isn’t being used frequently to displace top animators at the moment, Leinonen acknowledged that there is likely potential for generative AI to displace workers where clients are in search of quick, low-budget work, particularly among small studios. “If it’s going to make its way in, it’s going to be to get around low budgets.”

The displacement in the tech sector has yet to spread to other administration-heavy industries in Canada.

Graham Mitchell, organizing director for SEIU-West, says there are “lots of articles about how it’s going to be used in the diagnosis process, especially in rural and remote health,” but his union isn’t currently fielding a lot of grievances from workers about AI yet.

He notes that the labour movement tends to be reactive, with workers far more concerned about immediate issues like wages and safety. Additionally, he notes that his union often already negotiates language “around technological change and automation that prevents employees from losing work in favour of automation.”

Labour’s Role in Training AI

Focusing on worker replacement obfuscates the role of labour in the development of the large language models (LLMs) that power this type of AI. The AI models we’re familiar with (ChatGPT, Gemini, and others) have been trained on and developed by data input which hasn’t been categorized and labelled by crowdsourced labour.

Exploiting human labour to advance technology for further exploitation is surprising to precisely no one who has ever examined the historic development of class society.

AI is made up primarily of machine learning algorithms that are trained through interaction with people to develop humanlike responses.

A prime example (no pun intended) of this type of exploitation is Amazon’s MTurk service, where workers can create their own minute-by-minute flexible work schedule. The fact that one study found the median pay for MTurk workers to be only around $2/hour is indicative of the ways in which it allows hundreds of corporations to skirt minimum-wage and fair labour practices to exploit thousands of people in desperate socio-economic circumstances.

Exploiting human labour to advance technology for further exploitation is surprising to precisely no one who has ever examined the historic development of class society. All machines that have led to automation have been built to mimic replicable motions made by workers and have been constructed by funds produced through exploiting the surplus value produced by workers in the labour process.

In essence, the surplus value created by the labour power of workers has always been co-opted by capital to develop the very means of displacing those workers and further stratifying wealth.

The Impact to the Environment

One of the other key components of the AI discussion is its impact on the environment. Beyond just the massive raw material requirements for building AI data centres, there are also major concerns around the ongoing water requirements to run these facilities. The amount of heat generated by processors requires a massive amount of water for cooling.

According to some reports, a ChatGPT request consumes ten times the amount of electricity of a Google Search. Globally, this AI infrastructure may soon consume up to six times more water than Denmark, which currently has a population of 6 million people.

“We do find that the environmental aspect is often under-discussed,” Eddy Pedreira, president of IATSE Local 938 tells me. While he considers the climate impact as existing beyond the purview of collective bargaining, he notes that “our membership is very engaged with the environmental aspects of it.”

The divide between labour activists and environmental activists over harmful new technologies has a long history in Canada – right back to early logging practices, which were viewed differently across these spheres. Forestry workers wanted to flex their collective muscle to protect their working conditions and pay while environmentalists wanted to protect old-growth forests from being decimated. The most famous example of this is the “war in the woods” that took place in B.C. during the 1980s and 1990s, when New Left climate activists clashed with old-school union activists over clear-cut logging.

This division has been a setback for movements in the past, but Pedreira believes it may provide a path forward this time around.

The capitalist class is also skilled at framing public perceptions of worker resistance negatively: the Luddites are set up as an object of scorn and dockworkers seen as standing in the way of essential progress.

“We’re seeing a lot more angry, politicized folks,” he says. “A lot of people who want to take that anger and experience and put it into the movement.”

He envisions a future where labour and environmental activists unite to challenge the crushing momentum of technologies that are harmful to both workers and the environment.

The Impact of Technological Change

Historically, technological innovation has sparked fear and opposition among workers. The Luddites smashed mechanical looms; dockworkers struck in protest during the shipping container revolution. There are countless stories of new technologies being resisted by workers, only for that technology to be introduced anyway since the capitalist class can profit from it.

The capitalist class is also skilled at framing public perceptions of worker resistance negatively: the Luddites are set up as an object of scorn and dockworkers seen as standing in the way of essential progress.

What these characterizations fail to acknowledge is that, despite ostensible economic development, technological automation has historically displaced workers, depressed wages, and concentrated wealth even further with the ruling class.

The trouble is that it’s unrealistic to fully stop the advancement of this sort of technological change in its tracks (as the Luddites found out). Still, unions have often attempted it in the past and, at many times, succeeded in negotiating contracts that mitigate the impact of automation by guaranteeing retraining and preventing job loss.

Theoretically, tools that allow for the automation of certain aspects of work should provide a net benefit for society. In 1930, John Maynard Keynes famously predicted that technological innovation would lead to a 15-hour work week.

The trouble exists, as it almost always does, in the relationship between classes. As long as the tools of automation are owned by a ruling class and operated by a working class, they are unlikely to be put to use for the benefit of society, and the benefits accrue disproportionately to the class that holds ownership over them.

And what about unions using AI themselves?

It is axiomatic that all class struggle involves fighting a ruling class with greater access to resources and technology. Amazon and other corporations are unafraid to use AI tools to break worker solidarity through strategic scheduling and to spread anti-union messages. 

Another common tactic of large employers in provinces with short ratification timelines is to withhold information from unions for an extended period, only to dump all of it at once. Organizers end up sifting through pages of useless data, calling numbers of workers who are no longer with the company, and wasting time. 

“What’s the responsible use for this from the union’s perspective?” Gniposky asks. He acknowledges that his union is considering how they might be able to make use of these tools for good while making sure to “fight off the bad.

SEIU, one of the largest bargaining unions and external organizing teams in one such province – Saskatchewan – has considered ways in which these tools might help them succeed in large campaigns. It is possible that AI could help work through these data dumps to neutralize this particular employer tactic. Anyone using these tools, however, must also take into account the privacy concerns around feeding data into AI models that will consume that data for training.

If they were to use these tools, the value would be in “processing a huge swath of information quickly to help in the decision-making process,” Mitchell says. But, he notes, the province’s three-month span for collecting signatures in order to unionize and achieve legal ratification means that organizers here focus on smaller bargaining units and as such, there haven't been many cases where SEIU has needed the support of AI tools. This also means that the union hasn’t yet had to reckon with the privacy concerns inherent to AI usage.

Though, based on the conversations I have heard from organizers, most unions, while discussing AI use, are hesitant to fight back with the same tools.

“What’s the responsible use for this from the union’s perspective?” Gniposky asks. He acknowledges that his union is considering how they might be able to make use of these tools for good while making sure to “fight off the bad.”

As offices turn to AI tools for crafting communications, Gniposky believes that, when it comes to organizing, what’s most important remains “our ability to talk to [workers] in a face-to-face way.”

Ultimately, unions are handling the many concerns around AI the way they have handled most things in history: trying to get ahead, negotiating the language in contracts, and preparing their members for collective action.

Despite the doom and gloom around these technologies, Gniposky is hopeful about the road ahead.

“If I could convey one key piece of information,” he tells me, “it’s that in the face of all this stuff, we’ve organized more workers than ever before. Workers are winning. It’s fearful, it’s scary, it’s bad. And we’re still able to win.”