Earth Day 2025: SustAInability in Research

Artificial intelligence (AI) and machine learning are powerful tools that have significantly advanced cancer research. For Earth Day this year, Rosalind and Morris Goodman Cancer Institute (GCI) trainees and GCSS Green Labs Initiative members, Marie-Ève Proulx, Stephanie Han, and Kayla Heney, shared eye-opening insights into the environmental impacts of AI, and explored ways we can make our AI usage more sustainable.

 

The Dark Side of AI: Environmental Impacts

Massive amounts of natural resources and energy are required to train machine learning and large language models (LLMs). Researchers estimate that training ChatGPT-3 used resources equivalent to the energy needed to power 88 Canadian homes for a year (1287 MWh), and it produced the same amount of carbon dioxide as flying from Montreal to Australia 33 times (550 tonnes). The significant use of energy does not stop once the models are trained. Recent estimates suggest that each ChatGPT query can consume the same energy as running a 7 W LED bulb for 1 hour, and the energy used daily could power 18 Canadian homes annually (260 MWh/day). Energy consumption and carbon dioxide emissions are not the only contributors to the carbon footprint associated with AI and its use in research. Large amounts of water are required to cool the data centers and servers running these models, and the explosion of AI is leading to a rapid increase of electronic waste, which can contaminate soil and water systems if improperly disposed due to their hazardous materials, including lead, mercury and cadmium. Importantly, according to OpenAI researchers Dario Amodei and Danny Hernandez, the amount of computing power used for Deep Learning Research, a subset of machine learning, has doubled every 3.4 months since 2012, implying that the energy requirements for AI are not decreasing any time soon.

 

Making the Grass Greener: AI Use in Sustainability Research

As the popularity of AI increases and the environmental impacts become clearer, Green AI is emerging as an approach used to develop tools for sustainable machine learning, notably by improving hardware efficiency, developing sustainable data management methods, optimizing algorithms, and employing smaller models with lower complexity. Green AI also aims to use LLMs to tackle specific environmental issues, including developing predictive forecasting algorithms to optimize the production of renewable energies, to track iceberg movement and the rate of melting or to monitor global deforestation, which help focus research efforts on imminently threatened areas with formidable speed and accuracy. AI is also a useful tool to improve sustainability in manufacturing and energy production sectors, with some projects efficiently tracking and analyzing emissions using satellite images to provide a comprehensive view of the industry’s carbon footprint, or by assisting with accurate material detection of textiles to improve sorting and recycling efforts in the fashion and manufacturing industries. 

 

Conserving Today, Powering Tomorrow: Sustainable Considerations in AI Use

1. Calculate the carbon footprint of your work: There are available online tools available to monitor or estimate your energy consumption when using or employing AI. Notably, PowerTop is an open-source Linux-based system that monitors and analyzes computational power usage in real-time, while Green algorithms provides a free online calculator to estimate the impact of any computational algorithm.
2. Consider your electronic waste: Do you need to upgrade to the latest smartphone, tablet or computer? Can you repair your electronics instead of replacing them? Can you purchase refurbished instead of brand new? Have you properly disposed of your electronic waste once it’s reached the end of its lifecycle?
3. Increase your code efficiency: when possible, try to limit memory requirements, use optimized libraries for AI training, use updated software to decrease computational requirements, use code profilers to find areas for optimization, test work-in-progress on a subset of data to decrease the runtime and carbon footprint requirements during optimization.
4. Be mindful about your personal AI use: Do you need to employ AI for this particular task? Are there other approaches that will provide you with the same outcome?

 

AI such a powerful and versatile tool. We are not suggesting you completely stop using it today. Instead, the next time you open ChatGPT to help you write an email or summarize text, think about the energy costs associated and decide whether asking AI is the right approach for the task.

 

This article was written without the use of ChatGPT or any AI tools.

 

About GCSS Green Labs Initiative

The GCI Green initiative is led by students, researchers, and staff to make research at the Rosalind and Morris Goodman Cancer Institute more sustainable.

Our goal is to reduce the GCI’s environmental footprint by encouraging greener practices in the lab. With projects funded by the McGill Sustainability Projects Fund and our close collaboration with the other Green Labs Initiatives (GCI), we are implementing new center-wide projects, as well as sharing information and experience on how to improve sustainability all around campus.

For more information please email us as green.gci@mcgill.ca.