Some of Google's datacentres are being cooled by AI, the search giant has revealed, with an algorithm controlling how fans and ventilation work to keep the equipment cool while using less power.
The project has been developed by Google's AI offshoot, DeepMind. It thinks the project could save millions of dollars of power savings and concurrently, reduce its carbon footprint.
DeepMind's datacentre cooling project uses the reinforcement learning type of AI, which works upon a system of trial and error to determine the most efficient way of cooling equipment, from a power consumption point of view.
Deep neural networks are fed information from the hundreds of sensors within data centre cooling systems every five minutes. This then predicts what different actions would do to its energy consumption, working out which would have the biggest impact on power. After safety verification, these commands are then sent to the data centre where they're verified and implemented again.
This type of reinforcement learning was also used by Google company DeepMind's AlphaGo robot that managed to beat professional Go players last year. Although it works completely autonomously, it can be taken over by a human if the action is thought to be too risky to the datacentre's operation or power consumption.
Previously, data centre operators were expected to manage the cooling system with AI recommendations, but it didn't implement the insights. After getting feedback from those managing the data centres on a day-to-day basis, Google decided to have the AI control the cooling system autonomously.
Google's AI control system is already providing energy savings of around 30 a month and has given the company the motivation to start testing it with other industrial applications to help tackle climate change "on an even grander scale."
