Artificial intelligences are often rightfully criticized for their energy consumption. Training and running large language models is resource-heavy, and many companies rushed into developing their own AI competitors without considering the environmental impact. While AI is not exclusively an ecological problem, it may also be part of the solution. After all, this is a problem we’ve been struggling with for way longer than AI has existed, and having someone capable of analyzing all the power consumption data can help optimize the technology. Energy and tech companies are now partnering up at the Open Power AI Consortium to make this a reality.
Electric Power Research Institute
It all started with the Electric Power Research Institute (EPRI), a nonprofit energy company operating since 1972. They were among the first to criticize the power consumption needed to utilize AI and have been among the most cited for it since then. Even in the research, EPRI argues that while the numbers are bad, relying on technology is still the way to go. Energy demand has increased exponentially even before AI, with computing speed, improved cooling systems, and cloud data centers keeping the numbers at bay until around 2018, when demand started growing faster than the tech sustaining it.
AI requests consume significantly more energy than questions on Google search.Source: EPRI / Vries, A.D. (2023)
— The Greek Analyst (@greekanalyst.bsky.social) 2024-11-22T18:40:54.265Z
The consortium
Through the Open Power AI Consortium, along with giants like NVIDIA and Microsoft, EPRI hopes to accomplish many small goals to fill the demand gap. This includes developing and maintaining open-source AI models, libraries, datasets, and other assets built on global resources and meant for sector-specific challenges. Different partners highlighted different parts, from costs and affordability to clean and reliable energy. At its core, the Open Power AI Consortium opens up to AI as a potential ally in environmental challenges.
One particular example is interconnection studies. The critical role of electrical grids means that any change needs to be planned. These studies, meant to analyze all facets of the changes, can take up to four years to complete, time we unfortunately don’t have if we want to catch up to the demand. The consortium aims to shorten this by at least 5 times using AI. It’s easy to say AIs are resource-heavy, and it’s true. But they are a technology that can be shaped however we want. In the hands of energy experts, this can result in the reversal of the role of energy from a destructive force for the environment to a positive one.
Photo credit: The feature image is symbolic and has been taken by Markus Spiske.
Sources: EPRI Press / EPRI Research / Marc Spieler (NVIDIA)
