Artificial intelligence can help us in many aspects – even in the need for environmental protection. Initially, there is a lot of computing power from server centers. However, they are often powered by coal-fired electricity.
Artificial intelligence (AI) is often referred to as the oil of the future. There is indeed a spark of truth in the intercepted metaphor. Similar to the past with oil deposits, the following now applies: Whoever has first developed and made use of the AI, for which the money also flows. Not only are AI models behind social media algorithms, they also select the right people for job interviews, they could control self-driving cars in the future. But before machine learning models become really intelligent, you have to feed them with a lot of data and train them. It sometimes requires millions of cycles to make a model usable. As much CO as five cars This costs a lot of computing power – and thus also a lot of electricity, which does not always come from renewable sources. To calculate a single model, up to 283 tons of CO could be emitted, scientists at the University of Massachusetts have calculated. That’s more than 10 American cars blowing up in their lifetime. For the experiment, the researchers run known AI models on a single graphics processor and multiplied the energy consumption by the total time required for full learning. However, the researchers started from the energy mix that is common in the USA. There, most of the electricity is still generated from coal. In Austria, the energy mix is greener: in this country, more than 80 percent come from renewable energies. But AI models are not always trained where they are researched. Companies and researchers often rely on cloud computing providers for compute-intensive tasks. Unsurprisingly, IT giants Amazon, Google, and Microsoft dominate the computing power business in the cloud. While the companies have announced that they will focus more on renewable energy, market leader Amazon in particular is reluctant to look at the cards, according to a 2017 Greenpeace report. At least half of Amazon’s electricity is said to come from coal and gas. But most competitors are also a long way from a green cloud.
In Northern Virginia, the area with the highest density of data centers in the world, the share of renewable energy is in the single digits. The Internet is now responsible for around ten percent of global energy consumption. As AI enters more and more areas, its share of electricity consumption will increase. Light in sight In the future, quantum computers are to solve the energy problem of AI. Instead of zeros and ones, these can represent states that cannot be explained by classical physics. The quantum computers, which are still in their infancy, are expected to make huge leaps in performance. The same task would then be solved with much less computing power, which ultimately means less energy consumption and emissions. Another hope is optical processors that work with light signals instead of transistors and cause less waste heat. A large part of the power requirement goes into the cooling of the processors anyway, the derived heat is often blown out unused. At Google, an AI of all places invented a system that allows data centers to cool more energy-savingly. Using data from thousands of sensors, an algorithm predicts heat generation at different points in the server farm and adjusts cooling performance to it. And Amazon is simply using the remaining heat to heat up its new campus.