The GSF endorses the Environmental Impacts of AI Act to ensure the development and application of AI is pro-planet and responds to the urgent call for action outlined in the United Nations Intergovernmental Panel on Climate Change's 2023 report.
Let's explore how the AI Environmental Impacts Act can foster a culture of greening software.
Acceleration of AI
The bill highlights a 2022 estimate suggesting ‘the number of computational operations being used to create each of the largest AI models is currently doubling every 10 months’. The proliferation of AI startups and increasing collaborations between tech giants and research institutions underscore a relentless momentum of AI development that shows no signs of slowing down.
Responsible Innovation
The nature of AI, especially generative AI, demands substantial computing power and energy consumption. For example, one study found that generating a single image consumes 1.35 kWh of energy, more than what’s required to charge the average smartphone. Another study found that a conversation with 20-50 interactions was equivalent to spilling a 500ml bottle of water.
AI energy use could balloon to more than 85-134 terawatt hours by 2027, as much power as the Netherlands consumes annually.
The AI Environmental Impacts Act recognizes findings that AI requires intense computing power, resulting in high energy consumption. In the pursuit of establishing clear standards for sustainability, the Act will help ensure that the development of AI tools aligns with the urgent need to safeguard our planet's health.
Greening the Software Ecosystem
The AI Environmental Impacts Act lays the groundwork for nurturing a culture of sustainability within the tech industry by encouraging transparency and accountability. The bill calls to specific government agencies to:
Mandate a study on AI's environmental impacts across its lifecycle
Convene an AI Environmental Impacts Consortium
Create a voluntary framework system for AI developers to report environmental impacts
Given our commitment to open collaboration and knowledge sharing, this Act aligns seamlessly with our ethos of collective participation in data-informed discussion on AI's positive and negative impacts.
Supporting Green Software Projects
The AI Environmental Impacts Act would benefit a few of our projects, including the Software Carbon Intensity (SCI) Specification, Impact Framework, and Patterns of Green Software Engineering. The SCI Specification and Impact Framework aims to improve how we gather accurate and recent environmental data across a software’s lifecycle to reduce its carbon emissions and environmental impacts. The Patterns of Green Software Engineering offer AI patterns for measurement and reduction.
A reporting framework for AI applications would support developers in calibrating interventions for carbon optimization based on larger, reliable, and up-to-date data sets.
Specifically, the Impact Framework, an opensource measurement tool that calculates software’s environmental impacts by converting observable data into environmental impacts, is heavily reliant on high-quality data. The Act would provide a consistent source of environmental data for actionable insights to help shape future policies advocating adopting green software principles and patterns.
Looking Ahead: Greening AI
The GSF remains focused on supporting actionable initiatives and developing tools to reduce AI's carbon emissions and other environmental impacts. In the coming months, we’ll be:
Monitoring policies that call for transparent, detailed measurement of software to determine its environmental impacts
Exploring how to expand technical content to include Green AI and Software Carbon Efficiency Rating for Large Language Models (SCER for LLMs) in Generative AI (GenAI)
Seeking greater understanding of the application of the SCI to specific AI use cases
For organizations eager to make a substantial impact, aligning with industry leaders such as Accenture, Avanade, BCG X, GitHub, Intel Corp, Microsoft, NTT DATA Corporation, Siemens, UBS, and other influential collaborators, we invite you to join our community.
This article is licenced under Creative Commons (CC BY 4.0)