[Paper Review] Unraveling the Hidden Environmental Impacts of AI Solutions for Environment

Anne‐Laure Ligozat, Julien Lefèvre+2 more|arXiv (Cornell University)|Oct 22, 2021

Energy, Environment, and Transportation Policies22 references4 citations

TL;DR

This paper proposes a life cycle assessment (LCA) framework to comprehensively evaluate the environmental impacts of AI for Green solutions, moving beyond narrow carbon footprint metrics. It identifies that current AI for Green research underestimates full environmental costs, including material use and indirect effects, and argues that LCA is essential to assess both direct impacts and potential rebound effects, especially under large-scale deployment.

ABSTRACT

In the past ten years, artificial intelligence has encountered such dramatic progress that it is now seen as a tool of choice to solve environmental issues and in the first place greenhouse gas emissions (GHG). At the same time the deep learning community began to realize that training models with more and more parameters requires a lot of energy and as a consequence GHG emissions. To our knowledge, questioning the complete net environmental impacts of AI solutions for the environment (AI for Green), and not only GHG, has never been addressed directly. In this article, we propose to study the possible negative impacts of AI for Green. First, we review the different types of AI impacts, then we present the different methodologies used to assess those impacts, and show how to apply life cycle assessment to AI services. Finally, we discuss how to assess the environmental usefulness of a general AI service, and point out the limitations of existing work in AI for Green.

Motivation & Objective

• To address the gap in assessing the full environmental impacts of AI for Green solutions, beyond just greenhouse gas (GHG) emissions.
• To evaluate the limitations of current AI impact assessments, which often focus only on energy consumption and carbon footprint.
• To demonstrate how life cycle assessment (LCA) can be systematically applied to AI services to capture direct environmental impacts across all life cycle stages.
• To highlight the risks of rebound effects and third-order societal impacts when large-scale AI deployment is assumed to deliver environmental benefits.
• To advocate for integrating LCA into AI research to ensure that environmental gains are not offset by hidden or indirect environmental costs.

Proposed method

• Adopting life cycle assessment (LCA) as the core methodology to evaluate the complete environmental footprint of AI services, including material use, energy consumption, and end-of-life impacts.
• Comparing AI-based solutions with reference non-AI solutions using LCA to determine net environmental benefits or drawbacks.
• Applying LCA to both attributional and consequential frameworks—especially the latter when large-scale deployment is anticipated to account for systemic changes.
• Identifying and analyzing key environmental impact categories beyond GHG, such as resource depletion, eutrophication, and toxicity, using established LCA indicators.
• Using existing tools and databases (e.g., Ecoinvent) to quantify life cycle inventory (LCI) data, while acknowledging data gaps—particularly for GPUs and specialized AI hardware.
• Proposing that the AI community collaborate with manufacturers to improve data transparency, similar to open science principles, to enhance LCA accuracy.

Experimental results

Research questions

• RQ1 What are the full environmental impacts of AI for Green solutions beyond greenhouse gas emissions?
• RQ2 How can life cycle assessment (LCA) be effectively applied to evaluate the environmental footprint of AI services?
• RQ3 Why do current assessments of AI for Green solutions systematically underestimate their true environmental costs?
• RQ4 What are the potential rebound effects and third-order impacts of large-scale AI deployment on the environment?
• RQ5 How can LCA be adapted to account for the systemic and indirect environmental consequences of AI adoption in real-world contexts?

Key findings

• Current AI for Green research significantly underestimates environmental impacts by focusing primarily on energy use and GHG emissions, neglecting material flows and indirect effects.
• Life cycle assessment (LCA) is a robust methodology capable of capturing a comprehensive range of environmental impacts from AI services, including production, use, and end-of-life stages.
• Even when LCA shows a net environmental benefit, this reflects only a technical potential and may not materialize due to real-world behavioral, infrastructural, or systemic changes.
• Large-scale deployment of AI solutions can trigger non-linear environmental consequences, such as increased demand for critical materials (e.g., lithium, cobalt) or reliance on fossil-fuel-based energy, which attributional LCA may fail to capture.
• The consequential LCA framework is necessary for large-scale AI applications, as it accounts for market and system-level changes, unlike the simpler attributional approach.
• Significant data gaps exist—especially on the environmental impact of GPUs and other AI hardware—limiting the accuracy of LCA, and the authors call for greater industry transparency to close these gaps.