Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence

Artificial Intelligence, Privacy and Security, Innovation and Economic Growth, Networks, the Internet, and Cloud Computing and Cloud Computing

Article Snapshot

Author(s)

Kate Crawford

Source

Yale University Press, 2021

Summary

Artificial intelligence (AI) relies on natural resources, low-cost labor, and data. The production of AI technology harms the environment. AI systems rely on low-wage workers.

Policy Relevance

Assessment of the costs of AI should include costs to the environment and to workers.

Main Points

  • Nonhuman systems such as computers are not analogous to human minds; AI is not really intelligent, and “machine learning” would be a better term.
     
  • AI depends on extracting energy and mineral resources from the planet, cheap labor, and data.
     
    • Seventeen rare elements are needed to make AI, such as lanthanum, europium, and yttrium
       
    • Mining is associated with local and geopolitical violence, bringing war, famine, and death.
       
  • Clean technology is a myth, and portrayals of cloud-based computation should factor in costs to the environment.
     
    • Data centers are among the largest consumers of electricity.
       
    • China's data center industry draws 73 percent of its power from coal.
       
    • Shipping emissions significantly affect the Earth's atmosphere.
       
  • AI systems rely on cheap crowd-sourced labor to do tasks that cannot be done by machines, such as assessing violent videos or hate speech.
     
  • Algorithms that monitor worker productivity may assign workers very short or very long shifts, whichever is profitable; workers are unable to predict their schedules, and the algorithm does not factor in this human cost.
     
  • As data extraction becomes more invasive and researchers access data without interacting with their subjects, the risk that research using human data will inculcate bias or do harm rises.
     
  • The processes for classifying people by race and gender used in training AI are beset with stereotypes and error.
     
    • People assume that race and gender can be detected by machines, but this presents political and social problems.
       
    • Some AI systems try to detect sexuality or criminality based on headshots from drivers' licenses.
       
    • Companies that set up classification systems operate with little oversight or public input.
       
  • Large-scale AI surveillance originated with national security services, but now extend to classrooms, police stations, workplaces, and other everyday locales.
     
    • Surveillance supports central control at the cost of other types of social organization.
       
    • Surveillance systems blur the lines between law enforcement and tech firms.
       
    • Surveillance will radically reshape civic life.
       
  • AI ethics should include consideration of the labor conditions of miners, contractors, and crowdworkers; we should consider the experiences of those harmed by AI, and the systems' carbon footprint.
     

Get The Article

Find the full article online

Search for Full Article

Share