On the Cultural Integration of Artificial Intelligence


2025

-----------------------

The cultural integration of artificial intelligence has thus far progressed like the interlacing work of a giant loom. What began as scattered, innocuous threads – wires looping through our homes, streets, personal lives, and institutions – is now pulling tight as the tapestry begins to form. As expected, the capitalist class looms over the working person who is left at the trough of information to boil over an etiquette of moral and existential confusion. It is important that we mitigate these confusions, and build stronger cultural tools by which to hold on to our collective capacities and means of production.

For the capitalist class, the emergence of artificial intelligence technologies have provided a powerful tool of coercion. Working people will be more inclined to cling to their jobs, accept lower pay, or even make allowances to moral thresholds, when faced with the threat of intense automation in all sectors of all industries. Artificial intelligence is a hugely interesting, massively compelling technology; but culturally it is being aimed squarely at the working person.

And the working person has reason to be concerned about this technology, of this threat of automation. Indeed, industry scale artificial intelligence systems are very much already here. While the manifold flaws in these systems can be identified, we must also appreciate that we have only scratched the surface of this paradigm. While people may consciously interface with AI systems like ChatGPT – and hold sentiments towards the technology – the fact is that industry scale artificial intelligence systems gladly interface with everyone who uses a phone or computer every day.

A simple illustration of the utility of these systems is found in intelligence operations. In decades previous there were very real bottlenecks in staffing enough analysts to pour over signals intelligence – satellite images, telephone calls, radar charts etc. – so as to extract useful and digestible information from oceans of noise. Now, layers of multi-modal artificial intelligence systems can and are doing this work constantly, precisely, and meticulously without a day off. These systems do not even need to be exotic to have a powerful effect.

I will not spend time detailing examples of industry scale artificial intelligence; that is not the purpose of this writing. It is enough to say simply: lavender and where’s daddy, skyfortress, flock, embodied ai, dark factories, alpha persuade, the electromagnetic battlespace, behaviors, prediction, synthesis, and whatever the hell Meta is doing. If these words mean nothing to you I can only encourage seeking them out, as many of these forces are acting upon you in your waking life.

What is relevant is that the systems being entrenched across the globe are laying the foundations for a potential mass automation of labor.

Even if at present this is mere threat, or tool of coercion, it is worth asking if it is enough to sit idly by and/or contribute to such an endeavor. This question becomes a sort of Pascal's Wager: should the working classes seek emancipation in a techno-labor-futurism by voluntarily – and with limited or no compensation – contributing to the development of AI systems? Or, should the working classes cease in their contribution, and take hold of the methods by which artificial intelligence systems extract and automate labor. Indeed, most are already choosing their answer to this wager without being aware that it exists; that every photo uploaded, every digital interaction, every captcha is a vote cast in ignorance.

There is a possibility that the carcinogenic impulse latent within the American capitalist mode of production will inhibit the rise of AI systems. That, upon extracting everything it can from its own naval, the capitalist system is more likely to implode than it is to produce for itself AI systems capable of providing value and profit.

I find that questions like this are where our cultural capacity fails to stir deep sentiment. The question is impossibly abstract. The individual has nothing to grab on to, nothing to seize; what is an ‘artificial intelligence’ is not universally known or agreed upon. We are unable to organize our collective capacities around artificial intelligence because we do not even agree about what it is.

This is the principal challenge I will pose, of labor in the moment or age of AI. Before we decide what to do about artificial intelligence we must first agree about what it is, and prioritize understanding the architectures that make it work, and the powers we have over them.

The way we talk about AI must be based on a shared understanding of what AI actually is. We must stop approaching conversations about artificial intelligence through the lens of our own individual instance of the ChatGPT free tier. Instead, we should cultivate a foundation of language and behavior which encourages expression through a lexicon that actually conforms to the thing itself. This is the starting point on which actions can then be organized.

For my part, I will do my best to initiate this understanding by inviting you into a laboratory of ideas with steel tables and buzzing machines.

------
[Dissection]

If we were to dissect the composition of a contemporary artificial intelligence system, we would extract a few descriptive word-organs of choice: objective, learning, data, compute. Ethics, we can hope, would run through all of these.
Each of these organs is foundational to the artificial intelligence system, without one of them it either can not function, or will be functionally artificially unintelligent.

We will investigate these organs as though we are dissecting a beast, observing with our scalpel how subsystems interact and connect to form a whole. Our aim is to build a shared language which might give us the capacity to manipulate elements of the creature at a later time. Moreover, we will dissect these organs with our lens focused primarily on the levers of control we—collectively—have over them, a lens of labor; a focus on our collective capacity to manipulate these systems.

Of the organs of a contemporary AI system, learning and compute are highly quantizable; they can be quantified and expressed in discrete mathematical values. Compute refers to the actual hardware, its silicon-electric architecture, scaling, parallelism, and power efficiency. Learning refers to the algorithmic and scientific design of a model, the procedural rituals, optimizations, and many layers of abstraction that are orchestrated to crunch a lot of numbers. Together, compute and learning are descendants of Science, they are analytic in the extreme and produce measures and can be discussed mathematically.

I have always enjoyed conversations with a true quant, a genuine logician, anyone who thinks beautifully in math. I like to imagine how they experience the world. Learning and compute are organs they play far better than I could attempt to explain.
So I will not attempt to explain these organs, there are a wide range of free scientific resources available on the open web to learn more about these critical entities. We should be encouraged to develop a shared elementary understanding of the processes of these organs. If nothing else I will stress that anyone who only discusses these components of an artificial intelligence system is themselves, as of today, truly only comprehending part of the picture.

Instead of examining compute and learning in detail, we will turn our dissection tray to the organ of objective. Objective is the goal of an artificial intelligence system. Objective is on the one hand teleological: what is the purpose of this system in the first place? Objective is the answer to why a model is being created, why it is here and what ends does it serve: for education, play, to experiment, to achieve an outcome, to serve and be subservient, improve, identify, predict, or achieve capitalist ends, etc. The teleological purpose is often initially dictated by the creators, or the funding enterprise, which initiates the birth of an AI system; though, as perhaps with all things, purpose is not a static end.

In this light, the objective organ of AI systems is not something which we collectively have very much control over. In fact, the teleological goal of an AI system may result in the goal of that system being at odds with the intentions or goals of people. If the goal of a system, the reason it was created, was profit extraction from its users or targets, then that system—and the human infrastructures around it—can and will be optimized to fulfill that purpose.

The objective organ is also a technical regulator enmeshed with the learning organ. The objective determines certain technical decisions that are made by the learning organ, it decides what will be optimized, incentivized, what will be measured, and how important certain functions will be. The system itself is evaluated in large part based on the contents of its objective. This is an important practice in the development of artificial intelligence, to give the system itself a goal towards which it can make decisions, take actions, and learn. In fact, the learning organ would be rudderless without the objective organ, and would be unable to decide which things were important. The objective is also the location of the “stop-button problem” in AI, where the goal risks being more powerful than the stop button itself.

The objective is the most elusive organ in the artificial intelligence system. It can also be very fragile, and can quickly malalign if placed in the wrong hands. For example, if a company working in artificial intelligence changes its corporate goal, but does not give care to the objective organ of its AI system, a capability gap can quickly appear and bleed. Events like this are not uncommon. Insurance denials, job losses, missed opportunities, and many abandoned projects have all stemmed from mishandlings of the objective organ. Human organizations that frequently rush or pivot will find themselves especially susceptible to this slippery nature; they may not take the time required to transmit their new goals into the objective organ of their AI system.

The objective is also tied to the goals of people who create and interact with it. I will quote Yanis Veroufakis “We train their algorithms, which train us, to train them, to train us, and so on, in a feedback loop whose goal is to shape our desires and behaviour” . The objective influences, and is influenced by, organizational incentive and permission structures, techniques of productivity, communication, character, style, measurement, interpersonal relationships, and many more cultural artifacts. It is not necessarily, as of yet, that artificial intelligence systems immediately realign themselves due to these human influences. Instead, the reward pathways (goals) followed by people who create an AI system, and the behavior of people interacting with that system, are synthesized into it over time. This synthesis occurs in the real, where artificial and human intelligence systems interact based on the incentives and goals of each participant.

While we collectively may not have much control over the objective organ of a given AI system, the way in which this organ is tied to the goals of the people who create and interact with a given AI system do give certain groups some control over it. Waymo offers a good example of this dynamic. People working at Waymo are incentivized by any number of things: income, a chance to make robots, or a desire for safety because they interact with cars on the road every day. The structure and culture of the company add further interpersonal and career incentives, which may influence the decisions about capabilities, optimizations, evaluations. The result of these incentives, goals, objectives, synthesize in the labor of these workers: in code, in wires, and wheels.

The behavior and goals of the communities which interact with Waymo vehicles can also augment that AI system’s objective organ. People who want to get from point A to point B may ride in the vehicles, contributing towards the company’s profit objective. But people in a neighborhood may also vandalize the vehicles, or might push their local government to ban their presence. This final example is important for understanding our lever of control over the objective organ: if a local community votes to ban Waymo from driving on its streets, then it has augmented the objective of driving in that area. This lever of control is usually situational, not something available to all people all the time, or is dependent on other factors like local laws, politics, community organization and so forth.

In encapsulation, the objective in an artificial intelligence system operates on many levels. The teleological objective – the goal and purpose of the system – is strongly entwined with technical learning, and human factors of varying degree can synthesize with the objective in wonderful, abstract, practical, or dystopian ways. However, the power a given outside observer has over the objective may be limited in ways similar to the limits of “voting with your wallet” , or perhaps to local acts of legislation, or vandalism. The capitalist class, corporations, and nation state actors are largely at the levers of power in terms of directly influencing the objective for most industry scale AI systems. Continuing to develop an understanding of the objective, and our collective power to influence it, will be important work.

Our collective power to directly influence the organs of computation and learning are minimal. Outside of preventing the construction of data centers, disrupting learning environments, or mandating governments to build or open source artificial intelligence systems for the public good – all of which may be righteous outcomes – these organs are inherently constructs of science and technical application. The organ of objective is one over which the general population has some lever of power; but this power is limited in that it is often an abstract synthesis of incentives and actions (by people directly working on these systems, or situationally by communities interacting with them) around which collective organizing may prove difficult. Moreover, as the Pentagon’s recent actions with the AI label Anthropic demonstrate, the ruling class has even greater control over the objective organ, and can directly force augmentations to the teleological or technical goals of a given system.

This brings us to the final organ, data, over which we collectively have the most power and direct control. The organ of data in artificial intelligence systems is a vessel with the capacity to hold anything which can be abstracted as a discreet electromagnetic signal.

To put this another way, data is an electronic imprint, anything that can be abstracted into digital representation, into 1s and 0s. Data can emerge from any sensor, any specimen, place, time, and even from the human mind. We can imagine a painting of a person walking through a snowy field, and videos of them from dozens of angles, use probes to monitor the person's temperature and bloodflow, scan them with radar and lidar and sonar, capture audio, take chemical samples, position them with GPS, we can simulate their most likely next movements, write about them, and collate our perspectives and sentiments and testimonies to capture this event. On and on it goes until we have electric footprints across a range of modalities not just capturing an event itself, but all references to and notions about the event.

Before contemporary artificial intelligence systems and machine learning practices, these electric footprints, digital imprints, were indexed and organized in a space akin to a library. We could search for, find, read each item of data like books on a shelf, generally immutable information contained in clearly distinct records. We could then reference and make use of these records in human readable ways which were repeatable, programmatic, or deterministic.

In contemporary artificial intelligence systems, data takes a different form. Data no longer functions as human readable information or fact: records organized neatly and videos, texts, or temperature gages all existing in separate sections of the library. Instead, the organ of data in the contemporary artificial intelligence system is not readily human readable, is not programmatic, not necessarily based in fact, and does not conform to library architecture. Data is the filament of hyperspace, and the electromagnetic battlefield, it might be better said that data is like memory, an electromagnetic imprint of something happening.

More technically, in artificial intelligence systems, data—electromagnetic memory—substantiates as a deconstructed electromagnetic signal which is not human readable information, and which is organized as a hyper-construct in a manifold of referentials. In more basic language, we might recall our imagined “person walking through a snowy field” example; in the organ of data, every source of electromagnetic signal relating to this event—regardless of its modality or validity—might be disassembled and captured like points of dew within a cloud of mist. In this cloud of digital mist every recording and imprint of the person walking through the field of snow is mathematically self organized into a series of micro-relationships which congeal and form structure. This is a crude explanation of what becomes the “token” or “parameter."

With these images in mind, the organ of data in the artificial intelligence system can be described as an organ of memory and synthesis. Like a nervous system it interacts with the substances of reality, holding and sending signals which are electrically compiled or retained, and creating formations onto which meaning can be ascribed.

There are two primary uses of the organ of data in contemporary artificial intelligence systems. The first use of the organ of data is for learning. In the training of an artificial intelligence system via machine learning techniques, electromagnetic signals are fed through the learning organ, which is rewarded or punished by the objective organ for its accomplishments. We can explain this process in simplified terms with the analogy of a young human touching a hot frying pan, wincing, withdrawing, and through this learning to avoid similar danger. In this analogy the nervous system has given the human a negative reward—pain—which helps the human connect the elements involved in the scenario to deeply learn how they interact, and can be manipulated, such that pain is avoided in future. The same principles apply in the interaction between the organs of data, learning, and objective in the creation of artificial intelligence systems: where heat, the frying pan, the hand, are all constructs of the data organ, and pain is a response of the objective organ attempting to guide the system towards a goal (of not being burned, or cooking etc.). While this is an extremely simplified explanation, it is this interaction of data and objective which are the primary instruments upon which the learning process in machine learning occurs.

The second use of the organ of data is in application, at time of inference while the system is running. Here, the data organ carries fresh stimuli: text prompts, sensor returns like photos and videos, whispers, music, spreadsheets, artificial smells, or any other electromagnetic signal which the system has been created to operate upon. It is worth observing this process in consumer-grade artificial intelligence systems; one can send a photo to Google Gemini and ask the system to describe the photo sonically, to output audio with spoken English. This multi-modal interaction is enabled by the organ of data, which can be designed to hold any source of electromagnetic signal without discriminating as to the instrument which produced it. At the time of inference, while an artificial intelligence system is running, the now learned imprints from the objective, and learning organs operate upon—via the compute organ—the fresh stimuli carried by the data organ. It is a process perhaps again analogous to the nervous system which receives incoming stimuli in waking life and passes these to the brain which reacts and operates upon them.

We have now dissected our way through all four of the primary organs in a contemporary artificial intelligence system, describing each as we go in simplified terms. But it is worth holding our attention on the data organ for a short while longer, to explore why it is the organ that we, collectively, have the most control over.

We will do this in two parts, first with a brief illustrative story, and then with a statement of facts. Together, we will end this dissection with information that we can refer back to, so that we can better talk about artificial intelligence with shared understanding. And so that we can better organize and plan our society around, with, or without these systems.

Our story of the organ of data transports us from the lab to the banks of the Nile in ancient Egypt. There, the sun sent photons—quanta of electromagnetic energy—to fields of wheat which synthesized and grew. People harvested this wheat, developing methods to carry and transport it down the Nile for storage, processing, consumption. A culture of information, writing, administration, extraction developed around this production of energy, and organization of a society formed along with it.

Thousands of years after the ancient Egyptians I found myself near the Nile in Egypt working with hundreds of people to create data which would be fed to an artificial intelligence system. I told this story, and spoke of cave paintings and hieroglyphs, as I always do when I teach a new group of people how to create structured data for machine learning. Data is not merely numbers of a spreadsheet, words on a page, or calculations in math class, but a construct of memory and consensus, of energy capture, and communication. We would create data for the machine not in the context of black and white, correct and incorrect, but as forming a consensus reality to be carried by the data organ and fed to the learning organ to achieve the objective at hand.

It is important to expand our appreciation for what data is in the context and ‘age’ of artificial intelligence. Data is a raw input, it is electromagnetic signal, energy, memory. More than that, data is a means of production.

Indeed, data is harvested in the Global North and processed at large scale in the Global South. Large corporations contract digital factories especially in Africa, Southeast Asia, South America where people work to transform raw data inputs into structured outputs by following policies and specifications of reality written by teams creating artificial intelligence systems and setting the objective. These policies, used to instruct workforces in a place like Vietnam about how to create structured data, are notable; a Large Language Models responses about the Vietnam War are formed from data sources largely in English, written in the United States or similar, and structured for LLMs via policies mostly established in the West. This is a challenge of the humanities, of culture, and is a singular demonstration of a process which is being applied to the full spectrum of areas and use cases involving artificial intelligence systems.

These processes described above, which are evolving and becoming more powerful as artificial intelligence systems improve, are also critical in understanding how these systems can contribute to the displacement or replacement of labor. By centralizing data as a means of production, one might begin to appreciate that its harvest, processing, and feeding to an artificial intelligence system can—if done correctly—allow that system to learn to automate a given task. A product like the Ray-Ban Meta Glasses yields raw harvestable electromagnetic signals of thousands of people performing an action, this data is processed and structured at scale, and fed back to an AI system which can learn to operate upon that action type.

We, collectively – as workers, students, teachers, as people of the global village – have more control over the organ of data than any other because we have the ability to control data as a means of production. We largely have the ability to control the data harvest, which is used in the training of AI, and we have some ability to augment data at inference, by augmenting the world these systems must interact with.

This becomes a very human problem, not a technological one. We are in charge of the input mechanisms, the activities of everyday life. We must not shy away from talking to our neighbors, to our friends and families, we must strengthen our bonds with those we love, for love is a powerful force.

We are human. And in this AI age we must care for every interaction we have with reality. We must honor this fact, and give it supplication.

Understanding the organs of artificial intelligence systems, sharing that understanding, and creating a common vocabulary which allows for this discussion, is an important first step. We should not ignore our collective power over the objective and the data organ, we should learn to speak to one another about them. It is not until that step is taken that we might have any hope of harnessing these forces. As much as the capitalist class has seized upon AI as an opportunity, so too can the cultural integration of artificial intelligence be an opportunity for the people.