Google decided to give an AI the control for cooling its large data centre by itself. Well not exactly, giving an AI full control for maintaining the optimal temperature of its data centres, data centres that hold thousand of server and that power everything from Google Search to Gmail to Youtube, well not exactly thought the algorithms runs independently, a person is present to manage it and ca intervene if the algorithms ends up doing something risky, what did you already forget those tens and hundreds of movies showing AI going rogue.
Where did it all start?
Well it all started when Google wanted to improve energy saving and reduce CO2 emissions to help fight the climate changes and to save a few million bucks, because who doesn't likes a bit of extra cash. Last couple of years Google was testing algorithms that learn how to adjust the cooling system in the best possible way- the fans, ventilation and other equipments. A previous working of the system was that it would make recommendations to the managers of these big data centres and then they would decide whether or not it should be implemented. All this lead to big savings in the cooling systems. Now, Google says, it has effectively handed control to the algorithm which is managing cooling at several of its data centres all by itself.
How does it work?
These big data centres are loud and hot and needs to be cooled. A snapshot is taking every 5 minutes by the system and the information is then provided to DeepMind's neural networks. The work of the AI is to evaluate which action is required and help maintain the temperature by minimizing the energy consumption. That's not all these actions run through a number of safety checks before being implemented as an added precaution, it's been trained to prioritize "safety and reliability" over performance and cost savings. Reinforcement learning, the AI utilizes this techniques which learns through trial and error. The good old learn from your mistakes theory but for AI. According to Google this system is already saving them 30-40% energy. It's also continuing to find new techniques to satisfy energy saving goals. The company adds that data centres might not be the only places that could benefit from such a system.
The growth so far
According to Google' the performance of the data centres has shown an increase of 12% to 30% in duration of nine months. In winters the system took advantage of the cold weather to produce "colder than normal" water, which reduced the energy required for cooling the datacentre.
Not the first time
This is not Google's first try to this system; in 2016 it implemented a system developed by DeepMind that provided recommendations to human overseers. In a test conducted by the company, it achieved a 40 percent reduction in the amount of energy used for cooling and a 15 percent reduction in overall power usage effectiveness, the ratio of the total building's energy usage to its IT energy usage.