Stories you may like
The Human and Environmental Impact of Artificial Intelligence
Artificial intelligence (AI) has emerged as a transformative force in the modern world, reshaping industries, economies, and everyday life. Its growth has been exponential. In 2024, investment for AI and cloud companies in the U.S., Europe, and Israel was projected to reach $79.2 billion, with generative AI firms accounting for approximately 40% of this investment. However, the development and deployment of AI entail significant human and environmental costs that are often overlooked. Many systems and supply chains essential to the AI industry are associated with questionable labor practices, environmental degradation, and ethically dubious data collection methods.
The Environmental Costs
In the context of the AI industry, several critical minerals are required for its continued operation. Cobalt, lithium, and rare earth minerals are essential for producing the batteries, processors, and data storage infrastructure that power them. For example, lithium extraction has significant consequences for regions like Salar in Bolivia, the world’s richest site for lithium. Other critical minerals, such as several rare earth minerals mined in China, generate toxic pollutants that contaminate surrounding ecosystems, exposing local communities to significant harm. In Baotou, Mongolia, the refining process for rare earth minerals has created a toxic lake filled with more than 180 million tons of waste powder from ore processing. Spanning over five miles, this lake exemplifies the environmental devastation caused by the extraction of materials critical for AI development.
AI requires a considerable amount of energy to process essential tasks. This high energy consumption is responsible for substantial carbon emissions. In particular, AI model training for large-scale models like ChatGPT and DeepMind requires massive computational resources supported by data centers. The International Energy Agency estimated that electricity usage by data centers could rise from 460 terawatt hours in 2022 to nearly 1,050 terawatt hours by 2026. Furthermore, the carbon footprint of training a single large natural language processing model can emit over 660,000 pounds of carbon dioxide, equivalent to the lifetime emissions of five internal combustion engine cars. As larger models are developed, AI companies must address their models’ energy inefficiency and environmental impact. Investments in renewable energy, data centers that are more efficient, and sustainable mining practices are critical to mitigating these costs. Without systemic changes, this rapidly expanding industry risks exacerbating ecological degradation, resource depletion, and climate change.
The Human Costs of AI Development
Although "artificial," AI relies on human labor for its functionality. AI models require human workers to label, check, and process the data. Companies often use services such as Amazon Mechanical Turk to outsource this work to individuals in economically disadvantaged regions. These “ghost workers,” slip through the cracks of AI development. They are underpaid, overworked, and unnoticed. Often, the nature of their work is psychologically harmful. Companies such as OpenAI outsourced data labeling for ChatGPT to workers in Kenya, paying them less than $2 per hour. These workers were tasked with reviewing and labeling extremely graphic and harmful content, including depictions of child abuse, murder, and torture, to train AI models to filter toxic material. As a result, many workers suffered severe psychological harm due to prolonged exposure to such content.
This approach perpetuates global labor inequalities, disproportionately affecting workers in regions with weak labor protections. Behind the veneer of AI as a technology of the future, much of its success hinges on vulnerable people who lack job security, benefits, or avenues for career advancement.
AI and Data Collection and Classification
The human costs of AI development extend beyond exploitative labor practices. The data used to train AI models often originates from sources without the explicit consent of individuals, undermining privacy and ethical standards. These datasets, while aimed at improving AI’s functionality, may perpetuate a harmful image of people and reduce human identity to rigid categories.
For example, ImageNet, one of the most widely used datasets in machine learning research, serves as a bedrock resource for AI classification while also exemplifying how such systems can mirror and amplify societal biases. Developed in 2009, ImageNet classifies more than 14 million images into nearly 20,000 categories. These categories range from objects like “chair” or “apple” to human descriptors such as “fireman” or “criminal.” While most of these labels are self-explanatory and harmless, the “person” category stands out as being outdated and reductive. This category included several offensive and dehumanizing labels such as “slut,” “loser,” and “redneck.” Furthermore, other categories relied on reductive and stereotypical representations of race, ethnicity, and profession.
These issues prompted significant public backlash, particularly after the creation of the ImageNet Roulette project in 2019. This application classified users’ images by ImageNet’s person categories. In response, the dataset’s creators reviewed and subsequently removed nearly half of the subcategories in the original person labels. However, this effort primarily addressed the most controversial categories and did not fundamentally resolve the underlying biases in the classification system. The story of ImageNet shows that classification is never neutral; it reflects power dynamics and assumptions about identity, deciding which traits and categories are valued while leaving others out.
Reimagining AI’s Future
The human and environmental costs of AI are deeply intertwined with its technological advancements. Behind the image of a technology of the future, AI-powered applications are powered by a complicated system with a host of labor, environmental, and ethical issues. Creating a more equitable future for AI requires a collective effort from policymakers and corporations. For the environment, the push for sustainable technological development must also be prioritized, with investments in renewable energy and environmentally friendly practices to mitigate the ecological toll of AI. Labor rights must be protected to ensure fair wages and safe working conditions for the ghost workers that make AI systems functional. Finally, strengthening data protection laws is crucial to ensure individuals have greater control over their personal information and how it is used. The future of AI should not only focus on technological innovation but also on being socially just, ecologically sustainable, and ethically sound.
User's Comments
No comments there.