AI’s Carbon Footprint Problem

Original article was published on Communications of the ACM – Artificial Intelligence

By Stanford Institute for Human-Centered Artificial Intelligence

July 9, 2020

Researchers at Stanford University, Facebook AI Research, and Canada’s McGill University have developed a tool to measure the hidden cost of machine learning.

The “experiment impact tracker” quantifies how much electricity a machine learning project will consume, and its cost in carbon emissions.

The team first measured the energy cost of a specific artificial intelligence (AI) model—a challenge because a single machine often trains several models concurrently, while each session also draws power for shared overhead functions like data storage and cooling.

The researchers then translated power consumption into carbon emissions, whose blend of renewable and fossil fuels varies by location and time of day, by tapping into public sources about this energy mix.

The tracker determined that some machine learning algorithms consume more energy than others, and the researchers have incorporated an easy-to-use tool into the tracker that produces a website for comparing the energy efficiency of different AI models.

From Stanford Institute for Human-Centered Artificial Intelligence
View Full Article


Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA

No entries found