I have noticed a rather popular trend to oppose AI in its entirety: not just bad applications of it, but to oppose the technology as a whole. I have even noticed this catch on in supposedly “Marxist” circles, despite this being an incredibly reactionary viewpoint bordering on Luddism.
Some people might just be mildly skeptical of AI, but others have outright AI Derangement Syndrome to the point of complete insanity. If they see “AI” anywhere they have a breakdown and go attack everyone involved. I have seen several social medial profiles I follow degenerate to nothing more than complaining about AI every second of the day in every tweet, searching out anyone who has anything to do with AI and attacking them.
Discussing modern technology with western Marxists these days often feels like walking on eggshells because you have to be careful not to say the word “AI” or people will lose their shit.
What is AI?
Historically, it has been used incredibly broadly, such a person describing the movements of an NPC in a video game as their “AI” even if it’s just a simple path-finding algorithm, and thus referred to pretty much anything automated. Although, these days, it more specifically tends to refer to a very specific kind of algorithm: artificial neural networks.
An ANN is somewhat loosely inspired by how biological brains work. You have billions of neurons and each neuron has many connections to other neurons with different strengths to those connections. Each neural connection can have its strength augmented or diminished, and this has an impact in how information moves through the neural network.
The human brain has roughly 100 billion neurons, but because each neuron also has many connections to other neurons, the total number of neural connections is much higher, closer to 100 trillion. Since each neural connection’s strength can be adjusted, you can think of it as if your brain has a 100 trillion little dials that can be adjusted and each affect how information moves through the brain. If you want to teach someone to read, well, there should be a very precise configuration of those 100 trillion dials that will be able to allow it to process visual information as an input that produces text information as an output.
The process of “learning” is really just finding out how to efficiently adjust those trillions of dials so you get the desired output from a given input. An artificial neural network based on a similar concept, you have virtual neurons each with neural connections to other neurons, and each connection is assigned a “weight” which is the strength of it. Each of these “weight” values for the neural connections are sometimes referred to as its “parameters.” Most modern AI does not have 100 trillion neural connections like the human brain because that would require a lot of computer processing, but many are in the hundreds of billions (DeepSeek R1 is 671 billion).
Let’s say you have a lot of training data, which is data that has both a “question” and an “answer.” For example, if you want to teach something how to recognize characters from a photograph, your dataset would contain many “questions” — which would be photographs — and the corresponding “answers” — the text it that is represented in the photograph. There are then various algorithms which can take training data and use it to adjust the ANN’s billions of parameters such that it “learns” the patterns, and then once it has learned it, you can feed it an entirely new “question” and it should be able to “answer” it correctly. For example, you could feed it a new photograph with written text on it and it could tell you what text is written there. This is known as optical character recognition and is a form of AI.
Why AI is an Essential Technology
Ludwig von Mises, in his famous essay denouncing socialism, condemned it due to “limitations of human mind” that would make it impossible for humans to centrally plan an economy as complex as human societies in his time. What Mises did not predict is that humans would one day build computers which could carry out calculations far faster than anyone could ever dream of.
However, there is still a major limitation with the traditional model of computation. If you want to automate some calculation, you have to sit down and write thousands of lines of if/else statements in order to precisely model the problem so that it produces a desired solution. This becomes increasingly more cumbersome the more complex the problem becomes to the point where very simple tasks we humans do on a day-to-day basis are just practically impossible to solve this way.
Again, let’s return to the example of optical character recognition. Let’s say I task you to write a program to convert a picture taken to a camera into the a list of any text contained in the photo, for example if the photo is of a piece of mail, it should be able to read the address written on the mail.
If you actually sit down and try this, you will quickly find it’s impossible. There are far too many variables to account for. Even the exact same hand-written text have very different visual information if it’s taken under different lighting conditions, from different distances, from different angles. No one has the same handwriting, and even the same handwriting taken by two different cameras can look different.
You’ll never be able to piece together an algorithm with a bunch of if/else statements to solve this problem. Yet, it can be solved by instead creating a learning machine which can just feed a ton of data into it and it will figure out how to adjust its billions of parameters on its own. You can feed it images of billions of different letters and the corresponding text (data initially produced by humans) and then it will find the patterns on its own such that if you feed it a new image it will be able to identify the letters within it.
This means that AI is effectively a block box. Once you have trained the artificial neural network on a dataset, you know if you ask it a “question” it may give you the correct “answer,” but in a sense you don’t really know how. If you try to peer into the ANN, it’s way too complicated to make much sense of (although there are some research papers on people who have tried). That’s both the beauty and the weakness of it.
It’s a beauty because it means we can build systems that are capable of finding ways to solve problems beyond our capability of actually directly understanding and programming them to do so. It’s a weakness because they are more prone to error given the AI is a blackbox you can only really evaluate it by trial-running it. You can give your OCR AI thousands of pieces of mail and verify manually it reads the text correctly each time, but you can never be certain on the one-thousand-and-oneth time it won’t spit out gibberish. (Sometimes this happens!)
The main reason AI is so important is because, without is, Mises ultimately wins in the end. We will have hit a roadblock in automation because certain problems simply would be impossible to automate because they are too complicated. I gave optical character recognition as an example, but even converting audio to text is something too complicated to do without AI. If you have ever written a text message on your phone by speaking it, you have used AI.
The fact we cannot solve problems like this because they are too complicated without the use of AI proves that without AI there are limits to how complex a problem can be before it’s impractical to solve just by writing out a lot of if/else statements in the traditional method of programming. Given that economies continue to grow in complexity every day, if you had socialist economic planning, there must also be a limit to the efficiency of such an economic system because the economy could only get so complex before it becomes impractical to actually plan any longer.
While Mises was wrong for his initial reason to believe socialism cannot work, without AI, he would be correct, but for a different reason. AI is necessary to go beyond the limits of what kinds of problems can be solved simply by coding them by hand.
AI is Making Huge Technological Breakthroughs
One example I have repeatedly used is optical character recognition. Did you know that the United States Postal Service is already largely automated by AI? Used to, they would need tons of buildings filled with employees who would be sent letters and have the job of reading the text on those letters and sorting them to where they need to go. However, with AI, you can use optical character recognition to read handwritten text automatically and thus also sort it automatically. USPS is pretty much an engineering marvel powered largely by artificial intelligence.
One problem that has plagued the medical science community for a long time is that of protein folding. I will not pretend to understand biology as I am not a biologist, but as I understand it, a protein a long string of chemicals which when activated, will fold into a particular structure, and the structure depends upon its chemical composition. Proteins can be so complicated that nobody has been able to develop a general algorithm to predict how they will fold… until the development of AI.
Generative AI has also made huge breakthroughs in the material sciences. Chlorine is a poisonous gas historically used in war, and sodium is a metal which spontaneously combusts on contact with water. Yet, if you combine the two, you get delicious table salt. It’s not always obvious to predict how materials will behave when combined together, and many scientists spend many decades combining different chemicals and molecules together to find a material with useful properties. It turns out you can feed an AI tons of information about the behavior of materials and their chemical composition, and then you can ask the AI for a material with specific properties and it will be able to predict how to synthesize it.
Some of the recent breakthroughs in nuclear fusion have also come through AI. In order to fuse together hydrogen, you need very high temperatures, and they create these temperatures using a super hot plasma. Yet, the plasma is so hot it will destroy anything it touches, so the engineers build a chamber called a tokamak that uses magnets to contain the plasma inside of it in a way where it does not touch any of the walls in order to prevent it from destroying anything.
One of the biggest challenges with this technology is actually stabilizing the plasma. The plasma is not a solid object so it is hard to control it. Not just hard, practically impossible, as it’s a chaotic system. If you lose control of it, it will touch the side of the vessel and destroy it. The only way to tame such a chaotic system is, well, with AI. It’s just too complex of a problem to solve by writing a bunch of if/else statements.
You can, of course, also put AI into robots and have them control machines and carry out tasks automatically. China has been implementing this for autonomous machines that can carry out infrastructure projects like resurfacing roads and building dams, and also using it for self-driving buses.
You have probably used AI yourself. As mentioned prior, if you have ever spoken to your phone and it responded (such as writing a text message by voice or using something like Siri or Amazon Alexa) then you have used AI. If you have ever used Google Translate or any other sort of translation service then you have used AI. If you have ever used a Snapchat filter then you have used AI.
Modern day smartphones will use AI to when processing photos taken by the camera. Most people expect to just be able to point their phone at something and take a picture. This is not how cameras work: they require tons of adjustments to make a picture come out look decent, and so they use AI to adjust these parameters automatically. If you have taken a photo on your smartphone then you have used AI!
It is silly to be opposed to AI in general (artificial neural networks). It is essentially a form of automation necessary to make scientific and engineering breakthroughs in many fields. To oppose it is to oppose technological progress, it is to oppose the development of the forces of production. I understand people who criticize some of its specific use cases as many companies shove it into their products for marketing purposes even if it doesn’t improve the product at all, and sometimes even makes it worse.
However, “leftists” who oppose AI as a whole — in its entirety and not just in specific use-cases — need to be educated on this topic, and if they still refuse to budge, need to be kicked out of Marxist circles, as anti-technology primitivism is reactionary and detrimental towards the construction of socialism. You cannot be a Marxist and an “anti-AI” activist with AI derangement syndrome who goes crazy every time the word “AI” is mentioned regardless of the context. This is just being opposed to historical progress, the exact opposite of Marxism.
The proletariat will use its political supremacy to wrest, by degrees, all capital from the bourgeoisie, to centralise all instruments of production in the hands of the State, i.e., of the proletariat organised as the ruling class; and to increase the total of productive forces as rapidly as possible.
— Manifesto
A development of productive forces which would diminish the absolute number of labourers, i.e., enable the entire nation to accomplish its total production in a shorter time span, would cause a revolution
— Capital
