Artificial Intelligence, Automation, and Class Struggle
Introduction
The International Monetary Fund (IMF) recently estimated that up to 60% of jobs in the U.S. – and 40% globally – primarily involve tasks that can be done by artificial intelligence (AI)-driven automation. Of these “AI-exposed” jobs, the IMF estimates that half could be easily replaced by AI.
It remains to be determined just how many jobs will be affected, and on what timescale that will happen. But one thing is clear: we find ourselves at the dawn of a new industrial revolution, and automation through AI is one of the defining issues facing the working class and labor movement today.
From the time of the first industrial revolution to the rise of digital computers and automation in the mid-20th century, to artificial intelligence and advanced robotics today, technological developments have always presented the same question: will the new technology be used to benefit the vast majority of people, or the wealthy few?
If used in the interests of workers, automation and AI could mean shorter work weeks and relief from back-breaking, mind-numbing work for most people, without loss in pay. But instead, companies always use technological developments to cut costs by eliminating or deskilling jobs and driving down wages. This is because the bosses and Wall Street bankers currently control how technology is used, and they are driven by competition to maximize profits at all costs.
As workers and organizers, then, the essential question is not whether AI and automation is inherently good or bad. Rather, the essential question – and the main battle we must wage – is over who controls the technology of production: the workers or the bosses?
It remains to be determined just how many jobs will be affected, and on what timescale that will happen. But one thing is clear: we find ourselves at the dawn of a new industrial revolution, and automation through AI is one of the defining issues facing the working class and labor movement today.
From the time of the first industrial revolution to the rise of digital computers and automation in the mid-20th century, to artificial intelligence and advanced robotics today, technological developments have always presented the same question: will the new technology be used to benefit the vast majority of people, or the wealthy few?
If used in the interests of workers, automation and AI could mean shorter work weeks and relief from back-breaking, mind-numbing work for most people, without loss in pay. But instead, companies always use technological developments to cut costs by eliminating or deskilling jobs and driving down wages. This is because the bosses and Wall Street bankers currently control how technology is used, and they are driven by competition to maximize profits at all costs.
As workers and organizers, then, the essential question is not whether AI and automation is inherently good or bad. Rather, the essential question – and the main battle we must wage – is over who controls the technology of production: the workers or the bosses?
What is Artificial Intelligence (AI)?
Artificial intelligence (AI) is a generic term that refers to software or machines that can take inputs from people or detect aspects of their surroundings and act in a goal-oriented manner. Like traditional automation, they are often used to improve the efficiency of labor by automating (parts of) the job. Unlike traditional automation, however, they are not restricted to follow a predefined routine or set of instructions. They can take in flexible, free-form inputs or detect unstructured environments and respond in a custom manner.
How is this possible? AI technologies today predominantly use machine learning techniques, which use statistical algorithms to extract patterns from large amounts of data, and can apply these patterns to similar problems, allowing these technologies to perform tasks even without explicit instructions.
As an example, let’s take the language model ChatGPT, which can answer questions, compose essays, and converse in a human-like manner. Under the hood, the ChatGPT 4 model consists of an estimated 1.8 trillion parameters, which transform one’s input query into a human-like response. To find the optimal values of these parameters, OpenAI took virtually everything that people have written and posted to the internet – essentially every article, letter, social media post, anything really – and adjusted the parameters based on those examples until the model could emulate what people would write. This statistical process is called “training.” Recent successes in generative AI can be predominantly attributed to increasing the amount of the training data.
It follows that ChatGPT, or any AI technology for that matter, isn’t actually “intelligent.” That’s really just a marketing term. The reason these models work so well is simple: through training, they extract the essential content from work that countless people have contributed, and distill it all into a statistical model capable of emulating those people’s work.
Fundamentally, this means that AI itself is built off of expropriating everyone’s labor. ChatGPT is built not simply through the labor of the scientists and engineers at OpenAI, but through the collective labor of everyone on the internet, including writers and artists who may have gone uncompensated.
In addition, it’s important to note that AI systems like ChatGPT, Gemini (Google), and Copilot (Microsoft), also have major environmental impacts. Undergirding every AI model are vast data centers full of computers that run the training calculations and that process every AI response. These calculations take a huge amount of energy and thus emissions: Google’s greenhouse gas emissions have increased by 48% since 2019, and data servers are expected to consume 16% of all U.S. energy usage by 2030. Moreover, the heat generated by the computers requires lots of water to keep the servers cool: Microsoft's water consumption increased by 34% from 2021 to 2022, and ChatGPT 3 guzzles 500 mL (about a 16 fl oz water bottle) for every 10 - 50 responses it gives; ChatGPT 4 likely uses more water.
All of this is to say that there are multiple dimensions to what is at stake with AI. Indeed, the question of who controls AI technology affects not just the future of work, but also bears a significant impact on mitigating the effects of climate change.
How is this possible? AI technologies today predominantly use machine learning techniques, which use statistical algorithms to extract patterns from large amounts of data, and can apply these patterns to similar problems, allowing these technologies to perform tasks even without explicit instructions.
As an example, let’s take the language model ChatGPT, which can answer questions, compose essays, and converse in a human-like manner. Under the hood, the ChatGPT 4 model consists of an estimated 1.8 trillion parameters, which transform one’s input query into a human-like response. To find the optimal values of these parameters, OpenAI took virtually everything that people have written and posted to the internet – essentially every article, letter, social media post, anything really – and adjusted the parameters based on those examples until the model could emulate what people would write. This statistical process is called “training.” Recent successes in generative AI can be predominantly attributed to increasing the amount of the training data.
It follows that ChatGPT, or any AI technology for that matter, isn’t actually “intelligent.” That’s really just a marketing term. The reason these models work so well is simple: through training, they extract the essential content from work that countless people have contributed, and distill it all into a statistical model capable of emulating those people’s work.
Fundamentally, this means that AI itself is built off of expropriating everyone’s labor. ChatGPT is built not simply through the labor of the scientists and engineers at OpenAI, but through the collective labor of everyone on the internet, including writers and artists who may have gone uncompensated.
In addition, it’s important to note that AI systems like ChatGPT, Gemini (Google), and Copilot (Microsoft), also have major environmental impacts. Undergirding every AI model are vast data centers full of computers that run the training calculations and that process every AI response. These calculations take a huge amount of energy and thus emissions: Google’s greenhouse gas emissions have increased by 48% since 2019, and data servers are expected to consume 16% of all U.S. energy usage by 2030. Moreover, the heat generated by the computers requires lots of water to keep the servers cool: Microsoft's water consumption increased by 34% from 2021 to 2022, and ChatGPT 3 guzzles 500 mL (about a 16 fl oz water bottle) for every 10 - 50 responses it gives; ChatGPT 4 likely uses more water.
All of this is to say that there are multiple dimensions to what is at stake with AI. Indeed, the question of who controls AI technology affects not just the future of work, but also bears a significant impact on mitigating the effects of climate change.
How AI in the bosses’ hands currently affects workers
The bosses would like us to think that the biggest risk of AI is the existential risks it poses for humanity: that a sci-fi-esque superintelligent artificial “general” intelligence will one day replace and destroy humanity. By focusing the discourse on false, long-term existential risks, rather than real and current risks to workers, they can distract workers from the ways they are currently using AI to further exploit them.
So how does AI actually affect workers in a capitalist economy? Driven by capitalist competition to maximize profits, corporations necessarily deploy new technologies to increase productivity to capture greater market share and drive down labor costs. Historical precedent has shown that AI in the hands of the bosses can lead to:
Job loss
Stagnant wages
Increased monotony, frustration, and deskilling of work
Suppression of worker and union power
So how does AI actually affect workers in a capitalist economy? Driven by capitalist competition to maximize profits, corporations necessarily deploy new technologies to increase productivity to capture greater market share and drive down labor costs. Historical precedent has shown that AI in the hands of the bosses can lead to:
Job Loss and Stagnant Wages
Most apparently and egregiously, many workers are at risk of being unemployed and displaced entirely by AI. While new technology has enabled steady growth in productivity over the last two decades, total employment in the private sector has in fact stagnated since 2000. While previously the industrial revolution and the introduction of machinery in factories dramatically shifted the composition of the US economy and labor market from being one based largely on manufacturing to one based on services (a move which simultaneously lowered effective wages) — an AI revolution has the capacity to affect a much broader range of jobs, including those outside of manufacturing.
As previously mentioned, the IMF estimates that 60% of jobs in advanced economies (particularly in the U.S.) are “exposed to AI.” Moreover, a 2023 report by OpenAI finds that generative AI models, and particularly large language models (LLMs) like ChatGPT, expose 80% of the U.S. workforce to automation. The report finds that jobs like translators, survey researchers, clerks, writers, and journalists, are particularly exposed to LLM automation. LLM systems can automatically translate source material, write stories or news pieces, or answer phone calls and schedule appointments.
We see these trends already playing out today. BuzzFeed News laid off 15% of its workforce, aiming to replace them with generative AI. Similarly, in an industry first, Morehouse College is now implementing generative AI-driven virtual teaching assistants (TAs) to conduct office hours, teach lectures, assess and grade students, and perform other classroom tasks. According to VictoryXR – the company that is developing the virtual teaching assistant technology based on OpenAI’s ChatGPT, and which is eying K-12 education as well – “Finally, you can have a TA that works 24/7 and doesn’t need a paycheck!” Meanwhile, in the logistics industry, advanced robots made by Dexterity AI, Agility Robotics, and Boston Dynamics can automatically pack boxes and direct them to where they need to go – and UPS is rolling them out.
To further maximize their profits, and because they control the technology and our wages, the bosses can unilaterally line their pockets with the wealth generated from the increased productivity due to new technologies, rather than paying workers more. As new technologies have enabled productivity to explode over the years, median household income has been stagnant since 1975. Overall, this has resulted in a staggering rise in income inequality.
As previously mentioned, the IMF estimates that 60% of jobs in advanced economies (particularly in the U.S.) are “exposed to AI.” Moreover, a 2023 report by OpenAI finds that generative AI models, and particularly large language models (LLMs) like ChatGPT, expose 80% of the U.S. workforce to automation. The report finds that jobs like translators, survey researchers, clerks, writers, and journalists, are particularly exposed to LLM automation. LLM systems can automatically translate source material, write stories or news pieces, or answer phone calls and schedule appointments.
We see these trends already playing out today. BuzzFeed News laid off 15% of its workforce, aiming to replace them with generative AI. Similarly, in an industry first, Morehouse College is now implementing generative AI-driven virtual teaching assistants (TAs) to conduct office hours, teach lectures, assess and grade students, and perform other classroom tasks. According to VictoryXR – the company that is developing the virtual teaching assistant technology based on OpenAI’s ChatGPT, and which is eying K-12 education as well – “Finally, you can have a TA that works 24/7 and doesn’t need a paycheck!” Meanwhile, in the logistics industry, advanced robots made by Dexterity AI, Agility Robotics, and Boston Dynamics can automatically pack boxes and direct them to where they need to go – and UPS is rolling them out.
To further maximize their profits, and because they control the technology and our wages, the bosses can unilaterally line their pockets with the wealth generated from the increased productivity due to new technologies, rather than paying workers more. As new technologies have enabled productivity to explode over the years, median household income has been stagnant since 1975. Overall, this has resulted in a staggering rise in income inequality.
Poorer Working Conditions: Increased Monotony and Deskilling of Labor
When a subset of the workforce is replaced by technology, the remaining workers will be allocated to the task of supervising the technology, ensuring it runs smoothly, correcting errors made by the technology, and doing intermediate tasks that the technology cannot automate. These tasks are often much more monotonous and require less skill than doing the work manually. We saw this happen during the Industrial Revolution: artisans with specialized skills in their craft are replaced by menial laborers performing repetitive motions at machines. The introduction of early machinery resulted in a massive increase in safety accidents. Analogously, in the generative AI era, tasks like writing and art are reduced to prompt-tuning large language models and validating their outputs. Rather than machines being tools to aid human workers, human workers become mere appendages to machines, having to work in rhythm with the machine. This is not an innate characteristic of the technology, but rather how the technology gets used under capitalism.
Eroding Worker and Union Power
As a consequence of the above three factors, under capitalism, technology is used to diminish worker power. Most directly, AI can be wielded as a threat against workers fighting for their rights. For example, when graduate workers at Boston University went on strike, the university attempted to use generative AI to evaluate papers and facilitate discussion and thereby replace striking workers. More broadly, the introduction of AI means that bosses can always use the threat of job loss and replacement by AI to force workers to work longer hours, in poorer conditions, and for less pay. Furthermore, by deskilling and eliminating jobs, technological advances ensure that the pool of available workers is larger and more exploited. Thus, when workers at one plant, factory, or office demand more rights, bosses have the option to simply replace them with someone else, or to automate away those jobs. Overall, these effects result in workers having less leverage to fight against exploitation.
Case Study: UPS
United Parcel Service (UPS) has already implemented automation at many of its newer facilities, and plans to expand implementation into more advanced roles in the near future. This plan, created by UPS’s CEO and board members, dubbed "Network of the Future," would see the closure of an estimated 200 facilities in order to consolidate package volume into newer automated hubs requiring far fewer workers. This automation is enabled by advanced robots, sorting machines, and optimization algorithms that all rely on AI.
The reason UPS is automating its logistics network through AI is simple: it’s to increase its rate of profit and market share by providing more (or the same) services with fewer workers and thus lower costs. Furthermore, reducing the number and leverage of unionized workers also affords the company more flexibility to respond to changes in market conditions. Within the capitalist system, such practices are required for UPS to remain competitive within the logistics industry and not lose market share to companies like Amazon, FedEx, or DHL.
UPS has utilized automation primarily for sorting its packages and route optimization. Many facilities use automated belt deverters that read and push packages onto their correct belts while loading and unloading trucks is done entirely by hand. Network of the Future plans to entirely eliminate manual labor within UPS warehouses by implementing end-to-end automation.
This process has already begun in steps with the most recent being the implementation of RFID chips into all shipping labels and soon RFID readers in each truck. With the latest RFID network, package cars would be able to detect when a package has been missorted into the wrong truck so it can be corrected before being dispatched. The next phase for UPS is to implement mechanical package handlers to load and unload trucks. These mechanical arms developed by Dexterity AI robots are similar to the ones used in auto plants that would move packages from conveyor belt to truck with the RFID network double checking its work.
While Network of the Future has partially been implemented, it is already showing its class character on the shop floor. The current iteration of the RFID network requires manual RFID scanning by management. This useful technology to reduce errors has mostly been utilized against workers as a way for management to harass and threaten them based on performance despite these common mistakes being corrected as planned by the new network regardless.
UPS also has been using machine learning to optimize package flow nationally away from non-automated hubs to automated ones, reducing the required worker labor hours by 10%. Notably, UPS used this system to divert flow to automated hubs in the lead up to the Teamsters’ 2023 near-strike in order to reduce workers’ leverage. The effects have continued long after the contract settlement, too, with most part-time workers struggling for hours.
The reason UPS is automating its logistics network through AI is simple: it’s to increase its rate of profit and market share by providing more (or the same) services with fewer workers and thus lower costs. Furthermore, reducing the number and leverage of unionized workers also affords the company more flexibility to respond to changes in market conditions. Within the capitalist system, such practices are required for UPS to remain competitive within the logistics industry and not lose market share to companies like Amazon, FedEx, or DHL.
UPS has utilized automation primarily for sorting its packages and route optimization. Many facilities use automated belt deverters that read and push packages onto their correct belts while loading and unloading trucks is done entirely by hand. Network of the Future plans to entirely eliminate manual labor within UPS warehouses by implementing end-to-end automation.
This process has already begun in steps with the most recent being the implementation of RFID chips into all shipping labels and soon RFID readers in each truck. With the latest RFID network, package cars would be able to detect when a package has been missorted into the wrong truck so it can be corrected before being dispatched. The next phase for UPS is to implement mechanical package handlers to load and unload trucks. These mechanical arms developed by Dexterity AI robots are similar to the ones used in auto plants that would move packages from conveyor belt to truck with the RFID network double checking its work.
While Network of the Future has partially been implemented, it is already showing its class character on the shop floor. The current iteration of the RFID network requires manual RFID scanning by management. This useful technology to reduce errors has mostly been utilized against workers as a way for management to harass and threaten them based on performance despite these common mistakes being corrected as planned by the new network regardless.
UPS also has been using machine learning to optimize package flow nationally away from non-automated hubs to automated ones, reducing the required worker labor hours by 10%. Notably, UPS used this system to divert flow to automated hubs in the lead up to the Teamsters’ 2023 near-strike in order to reduce workers’ leverage. The effects have continued long after the contract settlement, too, with most part-time workers struggling for hours.
How Could AI Impact Workers if They Have Control?
Workers are the ones who best know what we need to effectively do our jobs, including what technology is most needed. In the hands of workers, automation could be used to create safer working conditions. At UPS, rather than being used to break strikes or cut hours without pay, package volume could be shifted to automated hubs during heat waves to reduce working hours during peak heat without a loss in pay. In addition, repetitive or strenuous tasks could be automated to prevent strain injuries. We could even use automation to reduce the work week without a loss in pay rather than laying off workers in droves – because with growing productivity should come wage and employment increases, not the other way around. We could also collectively balance the use cases of AI against its environmental impact.
Right now, AI is used to automate creative tasks that put workers out of jobs. In our hands, we could automate the monotonous and repetitive tasks so that we can continue to pursue the creative elements of our jobs. And we have seen examples of this already within labor union struggles, including most recently with Hollywood writers and actors.
Right now, AI is used to automate creative tasks that put workers out of jobs. In our hands, we could automate the monotonous and repetitive tasks so that we can continue to pursue the creative elements of our jobs. And we have seen examples of this already within labor union struggles, including most recently with Hollywood writers and actors.
Case Study: Hollywood Writers and Actors
The latest iteration of AI tools, known as generative AI, are particularly good at synthesizing text, images, and, more recently, videos. This means that they are geared to have a particularly outsized impact on writers and actors. Generative language models like ChatGPT have the potential to replace human writers. Meanwhile, generative video and generative image models (DALL-E, Midjourney, Stable Diffusion, Runway Gen-3 Alpha) can be used to create digital replicas of actors. Without strong labor protections, writers run the risk of being replaced with generative language models, being relegated to low-level “cleaning up” of text generated by large language models, and having the resulting productivity gains utilized against them in order to elongate their working hours and cut their pay. Simultaneously, actors run the risk of having their “likenesses” (e.g. their facial features, their voice, etc.) be used by production companies without their knowledge, consent, and without being paid. Background actors can be paid once for taking an AI scan, which then hands the rights of their likeness permanently to the studio, and can be reused over and over again in different movies. The latest video generators might dramatically transform or eliminate large sections of traditional film and television production jobs.
In April of 2023, the Writers Guild of America (WGA) went on strike. The following month, the Screen Actors Guild (SAG) also went on strike. Among both unions’ core demands was consolidating workers’ right to determine how AI is used. Following the conclusion of their strike, the WGA’s newest contract allows writers to “use tools like ChatGPT if they want to,” but these tools “can’t be used to undermine a writer’s credit or separated rights.” Writers cannot be forced to use these tools, and companies must disclose if they are handing scripts generated by these tools to writers. The SAG’s newest contract requires that companies request consent before taking AI replicas of actors; the companies must also disclose how those replicas will be used, and must compensate actors for their replicas. Building on those precedents a year later, workers with the International Alliance of Theatrical Stage Employees (IATSE) also fought for and just won a contract with AI protections in July 2024.
In April of 2023, the Writers Guild of America (WGA) went on strike. The following month, the Screen Actors Guild (SAG) also went on strike. Among both unions’ core demands was consolidating workers’ right to determine how AI is used. Following the conclusion of their strike, the WGA’s newest contract allows writers to “use tools like ChatGPT if they want to,” but these tools “can’t be used to undermine a writer’s credit or separated rights.” Writers cannot be forced to use these tools, and companies must disclose if they are handing scripts generated by these tools to writers. The SAG’s newest contract requires that companies request consent before taking AI replicas of actors; the companies must also disclose how those replicas will be used, and must compensate actors for their replicas. Building on those precedents a year later, workers with the International Alliance of Theatrical Stage Employees (IATSE) also fought for and just won a contract with AI protections in July 2024.
Class struggle unionism in the AI era
There’s a future in which AI and advanced robotics are used not to eliminate and deskill jobs, but rather to transform our lives as working class people for the better – to free us from back-breaking and mindless toil, and to enable us to have more time for our families, hobbies, health, and creative work. In order to realize that future, though, we will need to take control of that technology away from the bosses and bankers, and put it into the hands of workers.
From that perspective, the WGA and SAG strikes are critically important and inspiring reference points for today’s labor movement. Indeed, those unions boldly took on the bosses in a battle for explicit control over how AI technologies are used in production.
This is not unlike the Big 3 Strike of 2023, where the United Auto Workers (UAW) didn’t just fight to keep jobs in existing facilities, but directly and successfully challenged “management’s right” to open and close plants as they please. It is also notable that the UAW revived the concept of a 32-hour work week without loss in pay through that strike – raising the labor movement’s imagination and re-popularizing a bold working class demand that gets at the heart of who benefits from technology-driven productivity increases.
These are precisely the kind of forward-looking battles over automation that unions must continue to wage. But the magnitude of the fight goes far beyond what may be accomplished in any single contract fight.
Since the root issue is capitalist competition, what is ultimately necessary is to build working class power to effectively wage battle against the capitalist class entirely. This means aggressive new organizing in key economic sectors – for example, in logistics, organizing Amazon and Fedex workers alongside UPS Teamsters in order to fight the bosses over automation in that industry. It also means linking labor struggles with broader social movements of the day in order to unite the broadest section of the working class together in battle against our common enemy. In this regard, we can look for inspiration to the history of working class people uniting across unions and Unemployed Councils during and after the Great Depression to win victories that shook the capitalist system to its core.
We have a long way to go in order to build similar levels of worker power today. But if building such a movement was possible then, why not now?
From that perspective, the WGA and SAG strikes are critically important and inspiring reference points for today’s labor movement. Indeed, those unions boldly took on the bosses in a battle for explicit control over how AI technologies are used in production.
This is not unlike the Big 3 Strike of 2023, where the United Auto Workers (UAW) didn’t just fight to keep jobs in existing facilities, but directly and successfully challenged “management’s right” to open and close plants as they please. It is also notable that the UAW revived the concept of a 32-hour work week without loss in pay through that strike – raising the labor movement’s imagination and re-popularizing a bold working class demand that gets at the heart of who benefits from technology-driven productivity increases.
These are precisely the kind of forward-looking battles over automation that unions must continue to wage. But the magnitude of the fight goes far beyond what may be accomplished in any single contract fight.
Since the root issue is capitalist competition, what is ultimately necessary is to build working class power to effectively wage battle against the capitalist class entirely. This means aggressive new organizing in key economic sectors – for example, in logistics, organizing Amazon and Fedex workers alongside UPS Teamsters in order to fight the bosses over automation in that industry. It also means linking labor struggles with broader social movements of the day in order to unite the broadest section of the working class together in battle against our common enemy. In this regard, we can look for inspiration to the history of working class people uniting across unions and Unemployed Councils during and after the Great Depression to win victories that shook the capitalist system to its core.
We have a long way to go in order to build similar levels of worker power today. But if building such a movement was possible then, why not now?