---
title-long: "Addressing the Environmental Costs and Impact of AI"
title-short: Energy and Environmental Costs
document-id: environment
tags:
 - environment
# abstract in text format
abstract: >
  Developing and deploying AI systems requires natural resources (energy, water, rare earth metals),
  which puts pressure on energy grids, which are already under immense strain, as well as potentially
  causing harm to ecosystems and communities. There is a lack of meaningful evaluation of AI's
  environmental impacts and transparency around them; and a lot of greenwashing around AI's potential to “solve climate change”. 
# introduction and sections in HTML format
introduction: >
  <p>
    Recent generations of AI models, especially large language models (LLMs), have been using increasing amounts
    of compute; this is often presented in terms of GPU hours or millions of dollars used to buy cloud compute credits.
    However, this compute also comes with a cost to the environment; this can be conceptualized in a life cycle analysis (LCA) approach to AI.
  </p>

  <p>
    From the rare earth minerals that are extracted and transformed into computing hardware, to the energy used
    to power model training and deployment, and the water used for purifying layers of silicone and cool data centers;
    the usage of all these resources taxes planetary boundaries that are already under strain. Also, the concentration
    of power in AI means that a select few organizations are responsible for a large portion of the impacts, and yet the
    communities that are harmed by these impacts do not have a say in these; parallel with climate justice considerations
    that have already existed in the climate change space.
  </p>

  <p>
    Finally, while AI is part of the myriad of technologies that can help mitigate and adapt to climate change by improving
    weather forecasting, proposing new combinations of molecules for batteries, and even improving the efficiency of existing and
    future energy grids, it is currently unclear whether the environmental costs of AI technologies outweigh their benefits, and
    more transparency is needed to allow informed decision-making in the space. 
  </p>
sections:
  - section-title: Commercial and Infrastructure Effects of Energy Grid Demands
    section-text: >
      <p>
        Section text, HTML-formatted, TODO
      </p>
  - section-title: "Categories of Environmental Costs: Carbon, Water, Minerals"
    section-text: >
      <p>
        Section text, HTML-formatted, TODO
      </p>
  - section-title: Disparate Environmental Impacts and Cross-National Dynamics
    section-text: >
      <p>
        Section text, HTML-formatted, TODO
      </p>
resources:
- resource-name: Google Doc topic Card
  resource-url: https://docs.google.com/document/d/1fgi2NEb4glP8QGansroHmL9BZJ6Ja8zQKWQY4STRQ3k/
- resource-name: Primer on AI's Environmental Impact
  resource-url: https://huggingface.co/blog/sasha/ai-environment-primer
contributions: >
  Sasha Luccioni and Yacine Jernite wrote this document.