The massive size of datasets and the number of calculations needed to train Machine Learning algorithms generate a large (cloud-)server workload with a considerable carbon footprint. Therefore, to reduce the power consumption of AI applications, the European project SustainML has been launched. This multi-faceted endeavor seeks to create an innovative development framework to help AI designers from all ranges of experience.
In 2019, one study in the Journal Nature recorded a 300,000 kg of Carbon Dioxide emission from a typical training model used for natural language processing – the equivalent of 125 round-trip flights between New York and Beijing. As AI rapidly grows, the damage to the planet is increasing. Thus, the main goal for SustainML is to build a framework that would support AI designers by encouraging them to take energy conservation into consideration.
The project is being directed by the Spanish middleware business, eProsima, with further collaboration from IBM, French semiconductor business UPMEN, Kaiserslautern-Landau University (RPTU), Copenhagen University, German Research Center for Artificial Intelligence (DFKI), and Inria. Janin Koch from the Inria Saclay Centre, Université Paris-Saclay, and the CNRS within the LISN particularly focuses on how Human-Computer Interaction can be utilized to assist AI experts in making more eco-friendly choices throughout the AI life-cycle.
One direction the project is taking is the quantification of the environmental impact of algorithms. This helps point out the positive and negative effects of different decisions taken during the ML lifecycle. For example, opting to train an ML model in a cloud facility running on renewable hydroelectricity instead of a data center fueled by coal will make a large difference in the amount of carbon given off.
The SustainML project started in October 2022 with high ambitions to reduce the environmental harms of AI. This hopefully will provide a greater insight into the cost-benefit trade-off of decisions AI specialists make.
The company mentioned in this article, eProsima, is a Spanish middleware company. Across its short history, eProsima has provided a variety of targeted communication solutions and is becoming a leader in middleware solutions. eProsima specializes in providing high performance communication protocols and middleware for distributed applications.
The person mentioned in this article is Janin Koch. Janin is a scientist of the Ex-Situ project-team. Janin’s main expertise is related to Human-Computer Interaction and Artificial Intelligence, with a focus on raising awareness of the carbon footprint of Machine Learning and how Human-Computer Interaction can be used to get AI designers to make more sustainable decisions.
