- Sasha Luccioni, who is AI and Climate lead at Hugging Face, explored the environmental impact of AI models and focused on the intersection of AI technology and environmental sustainability, exploring ways to reduce the ecological impact of AI through better measurement and informed choices.
- AI models, especially large language models like ChatGPT, consume significant amounts of energy during both training and deployment.
- Inspired by the Energy Star program for appliances, Luccioni proposes a similar rating system for AI models.
Dr. Sasha Luccioni is a researcher in ethical artificial intelligence. Over the last decade, her work has paved the way to a better understanding of the societal and environmental impacts of AI technologies. Recently she explored the environmental impact of AI models and focused on the intersection of AI technology and environmental sustainability, exploring ways to reduce the ecological impact of AI through better measurement and informed choices.
1. Environmental impact of AI models
AI models, especially large language models like ChatGPT, consume significant amounts of energy during both training and deployment. This energy consumption results in substantial carbon emissions, comparable to those produced by several cars over their lifetimes. The training of such models can consume as much energy as annually dozens of American homes.
Also read: True Fit uses AI to help online shoppers find their best outfit
Also read: Sword Health raises $130M, valuation climbs to $3B
2. Development of energy star ratings for AI
Inspired by the Energy Star program for appliances, Luccioni proposes a similar rating system for AI models. This system is designed to help users choose models based on their energy efficiency, promoting more sustainable AI usage. The goal is to create a standardized way to measure and compare the energy consumption of various AI models across different tasks.
3. Initial testing and results
The project involves testing AI models on various tasks, including text generation, image classification, and speech recognition, using standardized datasets. Initial results show a wide range in energy efficiency among models, with some being significantly more energy-intensive than others depending on the task. These findings highlight the potential for optimization and efficiency improvements in AI model deployment.
4. Tools for measuring energy consumption
To aid in assessing the environmental impact, tools like Code Carbon have been developed. These tools estimate the energy consumption and carbon emissions of AI models, enabling developers to make more informed, sustainable choices. By using such tools, companies can reduce the environmental footprint of their AI technologies.
5. Future work and community involvement
The ongoing work includes expanding the rating system to cover more tasks and models, refining the measurement methods, and encouraging community feedback. The ultimate aim is working with organizations such as ISO or NIST for wider adoption to publish a Green AI Leaderboard to help the AI community compare and select models based on their energy efficiency, fostering a more sustainable AI ecosystem.
