By HERMAN MORRIS
Everywhere one looks in America, the message from corporations and even the government is clear. Artificial Intelligence (AI) is coming to change your life. Look at any screen and you will see how a new magical technology is going to change education, art, medicine, manufacturing, etc. It would be easier to list industries that AI isn’t being touted as a transformational tool. To back up these assertions, tech capital is making massive investments into AI infrastructure and commodities: $500 billion for stargate, a GPU datacenter for training AI models, hundreds of billions of dollars from Nvidia to develop more AI capable hardware, $65 billion for ai investments from meta. The investments into AI are so great that tech companies are investing into nuclear energy generation to meet the power demands of this new technology. Meanwhile, AI startups account for nearly half of all venture capital raised in 2024 according to Pitchbook. All this investment presents the question, what is society getting out of all of this? Is this going to transform the way everyone works? Is it just a bubble that is destined to fail?
What is AI?
It’s important to define what AI even is. While some form of Artificial Intelligence has been in use for decades now in the form of spell checkers, algorithmic content curation, and voice assistants, the current boom is largely around Generative AI models. In simple terms, these are computer programs that are fed large amounts of data (on the scale of the entire contents of hundreds of libraries), where they are instructed to identify and memorize the patterns found in that pool of data. After this “training” on data is completed, the models can then be prompted with queries in the form of photos, videos, or text and use its ability to recognize patterns from the prompts and data it was trained on to “generate” a response that appears wholly unique.
The importance of it being an appearance and not actually a truly original creation is key. While an ai agent can at times provide an uncanny ability to mimic human writing or image creation, it is doing nothing more than that, mimicking and pattern matching. This last detail is an inconvenient truth for the tech companies that are pushing these latest advancements in AI technology. Google has even gone as far as to fire scientists and suppress research at its company that demonstrate the limitations of Generative AI models.
Economic costs of AI development
It is also important to understand the cost of generating and operating these models. Historically, Large Language Models (a popular name for software that provides Generative AI functionality) such as ChatGPT or Gemini have extremely high upfront costs. GPT 4, for instance, costs over $100 million, and Gemini Ultra is estimated to have cost nearly $200 million. While information is hard to come by for operating costs, it’s known that OpenAI spent over 5.4 billion dollars a year and plans to increase its spending despite no path to profitability.
Despite spending the GDP of nation states to release and operate new models that, by these companies’ own admission, are giving diminishing returns, companies insist they are on a path to a higher level of AI technology that they call Artificial General Intelligence (AGI), so long as they keep getting more money. It helps them that AGI is such an ill-defined term, and that there is even disagreement amongst industry leaders on what it would do besides some vague sci-fi-esque intrigue on solving key world issues. Meanwhile, behind closed doors, companies like OpenAI and Microsoft are choosing to instead define AGI as any AI system that can generate more than $100 billion in profit. This is ironically just as fanciful a definition, given that no system generated yet has demonstrated a capacity for returning any profit at all.
On the other hand, there have long been signs that at least replicating the quality that the leading models produce could be accomplished with fewer resources. These fears were confirmed with the release of DeepSeek R1, an AI tool providing similar capabilities to Chatgpt while being trained on a fraction of the cost at $5.6 million. While OpenAI has alleged that the company illegally used OpenAI’s chat bots to train their model (a rich claim given that the best models available on the internet are only possible through rampant IP theft), so far the consensus is that the best models today can in theory and practice be mostly reproduced at a cheaper cost.
This would indicate that tech companies whose valuations have jumped on the promise of big AI features are overvalued by the market, as everything they have put out so far could be replicated by a business with a fraction of their capital. In turn, this points to an economic bubble that when popped will lead to layoffs, loss of savings, and misery for workers who may not even realize they are exposed to the economic failures of this sector of the economy. However, it also indicates that the current offering of AI commodities could be released and iterated on with a smaller budget and integrate with the economy as a real organ of capitalist production, as opposed to pure speculation.
How is Generative AI meant to be used?
So how are corporations trying to integrate AI into the economy today? While specific iterations differ quite a bit, broadly Generative AI technology in the U.S. seems to get applied in three ways—consumer facing technology meant to automate creative work or computer interaction, technology marketed directly to capitalists or the state itself to help surveil and discipline workers and people writ large, and lastly, a convenient smoke screen to allow for greater bureaucratic crackdowns on workers and the larger population.
The first category is what most people are likely familiar with. These are tools such as chat bots, text summarizers/rewriters, and image generators. These AI offerings are mostly offered for free with the option to opt in for premium features at a price. It should be noted that many of these creative tools are being deployed in the workplace in order to either intensify and automate labor to extract more value out of less workers.
As for monitoring software, there are examples from Israel using all the data they collect from Palestinian surveillance and using it to train a ChatGPT like tool to speed up their ability to collect and process monitoring of Palestinians. Additionally, startup companies are now offering surveillance technology to monitor workers in the office and factory line for productivity, automating detection and disciplinary actions for workers who are not efficient enough or not active enough at their job role.
Lastly, there are cases in which AI is being “deployed” by companies or state actors in a way that its stated value is a fiction and its real value is to provide a smoke screen for the most unjustifiable cuts and attacks on people (examples include AI denials of coverage from UnitedHealthcare and Elon’s use of AI to determine whom to layoff from the federal government). Any trivial audit of AI used in this manner would undoubtedly reveal at best a highly unreliable accuracy rate, and more likely, outright fraud to cover up the most unscrupulous attacks on workers.
This last category also captures attempts to replace wide swaths of workers in the federal government with the use of AI to streamline government services and surveillance. The new head of Social Security wants to use AI to detect fraudulent payments, and OpenAI is now offering a version of ChatGPT for the federal government. When these new efforts are paired with the massive cuts to the federal workforce, these expansions of the use of AI present a double threat, where pushing workers out of government jobs and the way of replacing them result in a worse system for managing the duties of the federal workforce. Additionally, these uses of AI will not come close to fulfilling their stated purpose of acting as an ideology-free bureaucrat as they will make it much easier to falsely deny services and to monitor anyone who relies on government services in their day-to-day lives. Instead, these efforts should be seen as part and parcel of the Project 2025 agenda for turning the government into an explicit weapon of reaction against anything they desire. This time, they have the bonus of a new fancy technology that they can hide behind and use to justify their attacks.
While none of these technologies have yet demonstrated a profitable business model, they should be watched closely for how capitalists expect to use these emerging technologies to speed up, automate, and keep closer tabs on productive labor in the coming period, and the DeepSeek revelations indicate that it might be possible for some of these technologies to have a life beyond the immediate tech bubble.
For workers’ control of industry and research
The above details demonstrate the dual threat that emerging technologies of all types present to working people. On the one hand, the ruling class plans to use them to further the rate of exploitation and efficiency of the capitalist system. And while there is always talk about how the workers who get thrown out of their jobs today will end up in new productive jobs tomorrow, the experience of workers whose jobs have been successfully deskilled or automated demonstrate that there is really no guarantee that this will happen. On the other hand, economic bubbles, even ones that form around productive technologies, pose just as much of a threat as misallocating the funding of the economy toward creating a painful correction that also throws workers out of production.
Both these risks point towards the importance of workers’ control of industry and establishing the right for workers to determine how new technologies are deployed in the workplace. Democratic worker-controlled funding for research could ensure that only uses that have been debated and voted on get approved to be researched. This could prevent the mad market scrambles that flood investments into a sector in a highly inefficient manner, chasing the most exciting and flashiest product rather than investing in actually possible uses of the technology.
Additionally, workers could discuss and vote on whether to introduce new technologies like AI into their field of work. And if it is judged to be possible to fully or mostly automate a job using AI, they could even decide on the terms of a just transition. We can look at last year’s strike of SAG-AFTRA and WGA for winning AI use provisions in their contract that allow for writers and actors to fully dictate whether they want to use AI in their work, with no loss of their own intellectual rights in determining how it is used.
While workers everywhere should look at these wins for something to replicate in their own workplace, it is important to go beyond union struggles and for workers to have democratic structures for funding the research and development of these technologies in the first place. Environmental considerations related to heavy power consumption, the extraction of rare earths and minerals, the excessive use of water resources, and increased waste and pollution should definitely be taken into account. At the same time, there are undeniable uses of these new technologies, especially in the medical sector where promising research breakthroughs for cancer detection are occurring. It should be up to workers to determine how social wealth is spent to invest in these new research ventures, and not the increasingly few capitalists who get to decide alone where the funds should go.
