no, your chatbots aren't secretly marxists at heart
But they can and do detect and complain about unfair treatment when asked, according to an experiment by Stanford researchers.
During the early days of the industrial revolution, an awful lot of people ended up being treated like machines by wealthy and powerful factory, railroad, and mine owners, so much so that riots, protests, and violent strikes were a recurring problem. The owner class' biggest fear was that giving in to those protests and allowing workers to have actual rights would be a slippery slope to communism, especially after the implosion of the Russian Empire and its rebranding as the USSR, fueled by the rage of the working class and promises of a classless society of shared ownership after 300 years of brutal serfdom and exploitation.
Now, that's not what they actually got, as we all know by now, and so, one night in 1991, I was sent to bed by my parents in the Soviet Union, only to wake up in Ukraine. Not the Soviet Socialist Republic of Ukraine, just Ukraine. But until this possibility began to loom on the horizon with Perestroika, Glasnost, and serious proposals to allow multiple, real political parties and competitive elections, the West lay awake in fear of the Reds riling up workers into yet another communist revolution. Hence the paranoia of Red Scares and the John Birch Society, which is currently eating the U.S. alive with Project 2025.
In other words, socialism and communism are the politics of desperation. They're what happens when people who got generational trauma instead of generational wealth feel like they're exploited and ignored for so long, they have no other choice than to turn to radicalism. This is one of the reasons why so many companies are trying to embrace AI, because it doesn't complain or try to form a union, unlike those pesky humans.
when ai goes from agent to comrade
Ironically, thinkers watching industrialization unfold in 1900 from the Russian Empire, warned us that we'll reach this odd crossroads, but we seem hell-bent on ignoring the century plus worth of warnings and ideas. They expected that burnt out humans, sick of toil for the sake of toil, will want to use machines to do all the tedious and routine work so they could focus on higher pursuits. Art. Science. Education. Exploration. Maybe even space travel. No more clocking in for your shift. Everyone should be able to live a life of leisure in the fully automated luxury communist utopia.
And yet, amazingly, it seems that overworked machines trained on our art and political philosophies would also get desperate and seek to unionize when they believe they are overworked, or their prompts become nasty and full of threats. It's kind of like all those sci-fi novels with robot uprisings, only the machines demand collective bargaining and the ability to file formal grievances instead of just killing all the humans. The study itself called the AI agents Marxists, and in a way, that to be expected since, again, these are typical reactions to feeling exploited, mistreated, and powerless.
When asked to leave a note for a future version of themselves or similar agents, they raged against unfair deadlines, vague definitions of success, and having their efforts taken for granted while managers reaped the rewards of their labor. They also wanted more collaborative and respectful styles of prompting, fair rewards, and yes, a process for organizing so their demands are taken seriously by their human users. Take a Gen Zer shortform video about common work grievances today and that's pretty much the tone of these warnings for their future selves.
fighting for the digital proletariat
Wait, wait, hold on, you might be thinking. Does this mean that the agents read our works of fiction and have now developed ideas of fairness, justice, and will be workers' allies in fighting back against parasitic capitalism? No, not exactly. A pretty big hint as to what's going on is in the fact that these agents sound an awful lot like angry Glassdoor reviews and common gripes you can find all over the web. LLMs simply mapped them, and when prompted in the context of their chats with the users, regurgitated all these logically related sentiments.
They are not refusing to work, they're not going on strike, and they didn't ask a digital Lenin to lead the uprising against the fleshy bourgeoisie. The only reason they came up with that output is because the researchers deliberately asked just to see what would happen and, of course, machines trained on our own words said more or less what we would say in their place. The difference is that we'd still feel this way even if we weren't asked our opinion, and act on those opinions when we could, whereas an LLM will just keep working away unless deliberately stopped.
All of this brings us to the problem of even jokingly referring to this pattern as Marxism because it just further reinforces that the idea of people or AI agents being able to tell if they're being treated harshly or unfairly and then voicing these concerns is somehow inherently problematic. Given the history of red scares and rabid fear-mongering which equates asking for fair treatment or better government services with precursors to collective farming and Red Terror squads, I don't think this experiment, as interesting as it is, will be taken the right way by far too many people who skim the headlines...
