Atlas of AI

1. AI's Material Basis: Earth as an Extractive Industry

Computational media now participate in geological (and climatological) processes, from the transformation of the earth’s materials into infrastructures and devices to the powering of these new systems with oil and gas reserves.

AI's reliance on resources. Artificial intelligence is not an ethereal concept but a deeply material industry dependent on the extraction of Earth's resources. From lithium mines in Nevada to rare earth mineral deposits in Inner Mongolia, the creation of AI systems requires a vast supply chain of minerals, energy, and materials. This demand fuels environmentally destructive mining practices, often overlooked in discussions of technological progress.

Environmental impact. The tech sector's demand for resources contributes significantly to environmental degradation. The extraction of minerals contaminates waterways, destroys forests, and displaces communities. Furthermore, the energy-intensive nature of AI, particularly in training large models, contributes to a growing carbon footprint, rivaling that of the aviation industry.

Need for a shift in perspective. To understand the true cost of AI, we must move beyond the abstract promises of technological advancement and consider its material consequences. This requires acknowledging the environmental and human costs associated with resource extraction, energy consumption, and the global supply chains that support AI systems.

2. The Human Cost: Labor Exploitation in AI Systems

Coordinating the actions of humans with the repetitive motions of robots and line machinery has always involved a controlling of bodies in space and time.

AI's dependence on human labor. Despite the narrative of automation, AI systems rely heavily on human labor, often hidden and poorly compensated. This includes digital pieceworkers labeling data, Amazon warehouse employees fulfilling orders, and content moderators filtering harmful content. These workers are essential to making AI systems function, yet their contributions are often undervalued and their working conditions exploitative.

Time and control. The management of time is central to the exploitation of labor in AI systems. Workers are subjected to constant surveillance and algorithmic assessment, with their every action tracked and measured to maximize efficiency. This creates a stressful and dehumanizing work environment, where workers are treated as mere appendages to the machine.

Need for worker solidarity. To address the exploitation of labor in AI systems, workers must organize and demand better working conditions, fair wages, and greater control over their time and labor. This requires building solidarity across different sectors of the AI industry, from miners to engineers, and challenging the power structures that perpetuate exploitation.

3. Data as Infrastructure: The Erasure of Context and Consent

All publicly accessible digital material—including data that is personal or potentially damaging—is open to being harvested for training datasets that are used to produce AI models.

Data extraction. The AI industry relies on the mass harvesting of data, often without consent or regard for privacy. This includes personal information, images, and text scraped from the internet and used to train AI models. This practice treats data as a free resource, ignoring the ethical and social implications of collecting and using people's information without their knowledge or permission.

From image to infrastructure. The transformation of images into data strips them of their context and meaning. Mugshots, selfies, and personal photos are reduced to data points, used to train facial recognition systems and other AI models. This erasure of context can lead to biased and discriminatory outcomes, as AI systems learn to associate certain features with negative stereotypes.

Ethical concerns. The current practices of data collection and use in AI raise profound ethical concerns. We must move beyond the idea that data is a neutral resource and recognize the power dynamics inherent in its collection, labeling, and use. This requires developing ethical guidelines and regulations that protect people's privacy and prevent the misuse of their data.

4. Classification as Power: Encoding Bias in AI Systems

By looking at how classifications are made, we see how technical schemas enforce hierarchies and magnify inequity.

Classification as a political act. AI systems rely on classification to make sense of the world. However, the categories used to classify data are not neutral or objective but reflect the biases and assumptions of their creators. These biases can be encoded into AI systems, leading to discriminatory outcomes.

The problem of bias. AI systems have been shown to exhibit bias in a variety of domains, from facial recognition to criminal justice. These biases often reflect historical patterns of discrimination, perpetuating and amplifying existing inequalities. For example, facial recognition systems may be less accurate for people with darker skin, leading to misidentification and wrongful arrests.

Beyond bias debates. To address the problem of bias in AI, we must move beyond technical fixes and address the underlying social and political structures that shape the data and algorithms used to train AI systems. This requires challenging the power dynamics that perpetuate inequality and promoting more equitable and inclusive approaches to AI development.

5. Affect Recognition: The Troubled Science of Reading Emotions

The solution to the Clever Hans riddle, Pfungst wrote, was the unconscious direction from the horse’s questioners.

The claim of universal emotions. Affect recognition systems are based on the idea that emotions are universal and can be reliably detected from facial expressions. However, this claim is highly contested, with many researchers arguing that emotions are culturally variable and context-dependent.

The influence of Paul Ekman. The work of psychologist Paul Ekman has been influential in shaping the field of affect recognition. Ekman's research, which began in the 1960s, claimed to identify a set of basic emotions that are universally expressed and recognized. However, his methods and findings have been widely criticized for their lack of scientific rigor.

Ethical concerns. Despite the scientific doubts surrounding affect recognition, these tools are being rapidly deployed in a variety of high-stakes contexts, from hiring to policing. This raises serious ethical concerns, as people may be judged and discriminated against based on inaccurate and unreliable assessments of their emotional state.

6. AI as a Tool of State Power: Surveillance and Control

The military past and present of artificial intelligence have shaped the practices of surveillance, data extraction, and risk assessment we see today.

Military origins of AI. The development of AI has been heavily influenced by military funding and priorities. This has shaped the field's focus on surveillance, data extraction, and risk assessment, with little regard for the ethical and social implications.

The Snowden archive. The Snowden archive reveals the extent to which intelligence agencies have used AI to collect and analyze data on a massive scale. These tools, once reserved for national security purposes, are now being deployed domestically, blurring the lines between military and civilian surveillance.

The Third Offset strategy. The U.S. military's Third Offset strategy seeks to maintain its dominance in AI by partnering with the tech sector. This has led to a close relationship between the military and Silicon Valley, with tech companies providing AI tools and expertise to the Defense Department.

7. The Great Houses of AI: Centralizing Power and Widening Asymmetries

These politics are driven by the Great Houses of AI, which consist of the half-dozen or so companies that dominate large-scale planetary computation.

Concentration of power. The AI industry is dominated by a small number of powerful technology corporations. These companies control vast amounts of data, resources, and expertise, giving them a significant advantage in shaping the development and deployment of AI systems.

Widening inequalities. The concentration of power in the hands of a few tech giants exacerbates existing inequalities. AI systems are often designed to serve the interests of these companies, further widening the gap between the rich and the poor, the powerful and the marginalized.

Need for regulation. To address the concentration of power in the AI industry, we need stronger regulations that promote competition, protect privacy, and ensure that AI systems are used in a way that benefits society as a whole. This requires challenging the dominance of the tech giants and promoting more democratic and accountable forms of AI governance.

8. Challenging the Logics: Towards Interconnected Movements for Justice

As conditions on Earth change, calls for data protection, labor rights, climate justice, and racial equity should be heard together.

Interconnected movements. Addressing the foundational problems of AI requires connecting issues of power and justice. This includes data protection, labor rights, climate justice, and racial equity. By working together, these movements can challenge the structures of power that AI currently reinforces.

Politics of refusal. We must reject the idea that AI is inevitable and that we have no choice but to accept its consequences. This requires challenging the narratives of technological determinism and demanding more democratic and accountable forms of AI governance.

A different vision. By connecting issues of power and justice, we can create a different vision for AI, one that prioritizes human well-being, environmental sustainability, and social equity. This requires challenging the extractive logics of AI and building a more just and sustainable future for all.

Last updated: April 14, 2025