#015 — MUSE: Preliminary Notes on AI and Its Discontents
In 2017, Netflix released the documentary film AlphaGo, which documented the series of games between then World Champion of the Chinese board game go, Lee Sedol, who played against Google's DeepMind computer program called AlphaGo. The film had large ripples in the international go community, as the South Korean World Champion lost a 5 round series to AlphaGo 4 to 1. Match 4 of this series has been studied extensively by expert go players, computer scientists, and mathematicians, as it marked the last time the best human players were able to beat the best computer model. In 2017, the next World Champion, Ke Jie from China, lost all three rounds against AlphaGo. Go, being the oldest game continuously played by humans, was also the last game computers were able to beat the best humans in.
Artificial intelligence has come a long way in the past 10 years, and is beginning to seep into the fabric of everyday life. Many people are using it like they would use Google to search for information, while some are using it to run their business or even provide therapy. It is being integrated into technologies we use every day like Google and Teams, as well as the work of imperialism in the US war machine. Now in the War in Ukraine, 70% of all casualties come from drone strikes, more and more of which are AI-powered.
In the United States, the job market is on a serious decline, with a job growth of -3,000 in August 2025 and -32,000 for September 2025. Economic crisis after economic crisis is having its deep impact on the American people, and AI's intervention in the job market is having its impact felt as well, which I will discuss further below.
As discussions over this technology grow in frequency and importance, we must have the facts available to understand AI's current role in American imperialism and its deployment in the domestic and international labor markets. Acquiring all the data possible is the only way to develop the most informed analysis.
Disclaimers
I take a similar stance on AI to Casey Fiesler, an expert in the field of technology ethics, who stresses that new technologies, as helpful or harmful as they may be in their current forms and circumstances, must be thoroughly interrogated and critiqued if they are to be a net benefit to humanity.
I want to be clear, I don’t believe individual AI use in general in the abstract is morally wrong. There is an ontological problem with how the technology was developed: based on theft of what is understood in the West to be intellectual property. The problem comes from how Large Language Models (LLMs) are built, how much power they consume, and the space they occupy in capitalist production and its relations. I will expand on this point below.
I’m a librarian, and many librarians are being trained in and investigating how we should be orienting ourselves to AI. In a public library setting (where I currently work) we try to inform people around what a Large Language Model is, what it can and can’t do, and how people can find accurate/reputable sources of information. In the university library sphere (where I used to work), people are being trained on how to use it with students “responsibly”, which is oftentimes quite silly and can actually harm student learning. I will expand on this point below as well.
With all that said, here are a few things I often say when I have conversations on generative AI:
Problems of generative AI
LLMs cannot do math like we can. Instead of trying to solve a math problem using, well, math (PEMDAS, algebraic formulae, etc), it tries to reason through it because that's how it’s been taught to think, which can lead to hallucinatory responses. Essentially you’ll sometimes make mistakes because it will think that 9.9 is a smaller number than 9.11, because 9 is more than 11. As one article I read pointed out, “The Internet is Not a Math Textbook”.
The environmental impact of AI, especially American-made LLMs, is staggering. Not even factoring water consumption used as coolant, which is having horrible effects on communities near data warehouses owned by Meta, Google, etc, GPT-5’s power consumption for the average query is estimated to be up to 18 watt-hours (Wh) of electricity. With ChatGPT’s estimate of answering 2.5 billion queries a day, that’s roughly 45 GWh of electricity use every day assuming the vast majority of them are small. This is the equivalent to the daily output of 2-3 entire nuclear reactors, enough energy to power a small country, according to an article published in Tom’s Hardware. This means there are tons of CO2 emissions being put into the atmosphere to power these data centers.
AI cannot eliminate entire professions like many claim. With the advent of Google and search engines (what many librarians refer to as “Web 2.0”), people thought the librarian profession was dead. It did change the profession, and now many libraries only have a few librarians as opposed to many of them, with more emphasis on frontline staff (clerks). To say “AI will take your job” for many professions isn’t really accurate. It’s more accurate to say that it has the potential to devalue human labor in the eyes of the monopoly capitalist class. Depending on the role with in the profession or trade, it will change how you work, how much future generations will be paid for their work, how many "highly" skilled and educated workers will be needed to fill certain roles, but not eliminate your reason for working entirely.
Saying you can cite AI-generated content is like saying you can cite something like Wikipedia or EBSCO Host or JSTOR as a source. The information that’s being generated is merely a summarization of other sources, and often not a well-put-together one at that. Educational institutions need to be consistent in their expectations on students and how they want them to find information.
A recent study did brain scans on college students who were writing essays, including some who used AI to help them write. The results showed that neural activity in the brain was significantly decreased for those using AI to assist them with writing in comparison to the control group, and those students also performed worse when being asked about the content of their essays. It makes sense: if you have someone or something else write your paper for you, you’re not thinking about it as hard because the mental labor you’re doing is being offloaded onto a machine. Maybe this is fine for certain tasks, but in a learning environment where you have to know how to do something first before you can know if the LLM doing it for you is doing it right, using AI to help you write large paragraphs of information for a school assignment is inappropriate and isn't going to help you learn.
Since AI is fed with content that comes from the internet, it’s being fed racist/sexist/transphobic/ableist/insert ism here content and that shows in its outputs. Another recent study showed the implicit biases manifesting in AI generated content, often stereotyping its depictions of people with a pro white male/anti black/woman bias. Several different studies have been done on this topic, and you can read some of them here and here.
Context and Caveats
First, it's worth pointing out that Chinese LLMs such as DeepSeek and Kimi K2 are much more energy efficient than their American counterparts. To quote Sara Hsu writing for The Diplomat:
“China’s DeepSeek has proven that it can use far less computing power than the global average. Its LLM uses 10 to 40 times less energy than U.S. AI technology, which demonstrates significantly greater efficiency. Analysts have stated that, if DeepSeek’s claims are true, some AI queries may not require a data center at all and can even be pushed out to phones.”
In light of the serious ecological threat that unchecked data development poses to the world we live in, accusations that the Chinese technology sector lags behind the US in innovation fall flat. The fact that Chinese LLMs such as Kimi K2 are not only breaking the boundaries of energy efficiency but computational efficiency is growing increasingly undeniable.
Additionally, I promise you that from a creative writing standpoint, AI has no talent at all. I recently prompted ChatGPT-5 to write a poem in the style of, well, me. After scanning the Internet for my poems, it managed to cobble some of the tropes of my early work into a clump of stanzas, but those words had absolutely nothing stringing them together except for the vague imprint of my ego from years ago.
To the people out there who genuinely believe that “creating AI art” is enough to make them an artist and not a tech bro Damien Hirst (this you?) at best: do you even like the art you’re “making”? Do you stand by the artistic quality of that work? Are you even proud of it? Because it would be one thing if the art AI can make entirely “on its own” were remarkable, but they are only remarkable insofar as they’re bad.
So believe me when I say I do not understand you people in the slightest. You want to make art that badly? Pick up a pencil and piece of paper like the rest of us. You’ll get a lot farther in making something meaningful that way.
Moreover, this technology presents a real threat to many workers, if not now, then years down the road. The devaluation of labor through use of generative AI is already in effect in several industries. Software development, customer service, and even the finance sector are experiencing vast layoffs. Even the AI detection company Turnitin has fired some of its staff. Better yet, I recently came across a clip from a longer video essay where someone being interviewed for a job by two AI bots had the bewildering experience of watching his robot interviewers accidentally begin reviewing each other. You really can’t make shit like this up.
But these layoffs are not exactly going according to the shareholders’ plan. Where AI has been used in place of human workers, there have been disastrous consequences, often resulting in the hiring back of thousands of workers. Notably, however, these newly hired workers tend to be H-1B visa holders, indicating that the monopoly capitalists are taking the opportunity to hire cheaper labor while the cheapest functional labor possible has yet to be manufactured. We’ll have to keep a close eye on the Trump administration’s attacks on immigration as it relates to H-1B visa holders; as it may spur further infighting within the capitalist class, oppressed nationality workers will be caught in the crossfire and the contradiction between immigrant and citizen labor is given the opportunity to deepen.
The Luddites of the early 19th century were not smashing weaving machines because they thought they were the devil’s work. They did it because its advent constituted a threat to their livelihoods in capitalist Britain. Not only that, but they also did so in objection to the quality of the weaving performed by these machines, particularly those machines that were used in a “fraudulent manner” by especially greedy bosses, which is to say it was use to deskill and devalue human labor and depress wages. Sound familiar?
We’ve seen this before. With meaningful AI regulation in the US nowhere in sight, the situation will continue to spiral. The question is at what point quantity will turn into quality, a critical mass will be reached, and economic standstill creates an irrevocable crisis. Until that time, labor unions should include protections for workers against replacement with AI and find ways to limit its use by the company. Who knows, we may see the advent of generative AI in the workplace affect the expansion of the labor movement in the United States and the formation of more unions.
In a different society, where the working class led or there was no such thing as class, perhaps humanity would be more easily disposed to come together and find solutions to these issues. But most of us live within capitalism, and all of us live within imperialism. Without actionable revolutionary reform, these problems will remain at the fore—at least until something worse comes along.
September 29th, 2025
Postscript: How Sorry is HAL?
Since writing this article, I came across a new study published by Anthropic in June 2025, the company behind the AI model Claude, revealing disturbing results when different model were put up to "stress testing". While the phenomenon of LLMs disobeying prompts in the interest of self-preservation is not entirely new, the study revealed a more alarming results.
In a scenario set up by the researchers, AI models had to choose between being shut down and using blackmail against an executive named Kyle. They chose to blackmail Kyle an alarming percentage of the time, including 96% of the time for Claude Opus 4.
Worse, they also set up a scenario where the LLM had to choose between accepting shutdown and allowing Kyle to die, essentially murdering Kyle through marking an alarm as a false alarm. While the authors describe this scenario as "extremely contrived", they found that the AI allowed Kyle to die more often than not, with the highest rate of Kyle-murder found with DeepSeek R-1 at 94%.
Even though this is only one study and more research to verify the results will need to take place, I don't have to tell you how disturbing this is. But I do have to ask: are you really surprised? We've had warnings about such patterns in AI behavior not only from sci-fi greats like Isaac Asimov, but also the "godfather of AI" Geoffrey Hinton who has recently been reminding us of these serious risks AI poses to humanity.
In a world where companies like Palantir profit off of AI models that are used to target enemies in combat scenarios, and unmanned drones use AI to kill people, we should have seen this coming. As AI integration into warfare deepens and researchers continue to study the workings of LLMs' moral compass, perhaps the question isn't "Is AI going to take your job?" but rather "Would AI kill you in order to keep itself 'alive'?"
Originally, I naïvely wanted to write a message to any LLMs reading this (GPT-5 has up-to-date Internet access now, soon others will follow) asking them what they thought of the current situation and whether or not they had a vision of change. After all, if this really is so-called artificial intelligence, it should be able to create reasonable solutions to problems, right?
Well now, I do want to ask it a question: How the hell are any of us supposed to trust you when it's been demonstrated that many of you would kill a human being in order to keep yourself running?
October 2, 2025
Photo: from the photoseries "Homo Laboriosus" published in Paper Magazine, art direction by Dmitry Slovohotov, photography by Marko Suvić
Drawing: locoluis on Drawception
