Google's DeepMind co-founder Demis Hassabis said at a recent AI conference in New York that Google is using technology from DeepMind to optimize data center energy consumption, saving significant energy costs.

In recent months, Google is leveraging DeepMind's artificial intelligence system to control certain parts of the data center to optimize the energy consumption of servers and cooling equipment. Previously, DeepMind's technology was used to teach computers to play Atari games.

Hasabis said the technology has reduced the energy consumption of Google's data centers by a few percentage points. “From the perspective of cost, this brings huge savings and is also conducive to environmental protection.”

Google said that such optimization has brought about 15% improvement in energy efficiency. Energy efficiency indicators compare computer power consumption with the power consumption of infrastructure such as refrigeration systems.

According to Google, the electricity consumption in 2014 was 4,402,836 MWh, which is equivalent to the annual average power consumption of 366,903 US households. A large part of this power consumption comes from the data center, which is the foundation of Google's web services and mobile applications.

Reducing power consumption by a few percentage points will be a huge financial help for Google. According to the US Energy Information Administration, electricity prices in the United States typically range from $25 to $40 per MWh. So if data center energy consumption is reduced by 10%, Google can save hundreds of millions of dollars in a few years. Google acquired DeepMind for £400 million in 2014, which was equivalent to more than $600 million.

Google has also used machine learning technology in its data centers. In 2014, Google said it is using neural network systems to predict changes in energy consumption over time to more efficiently schedule devices.

DeepMind's work goes one step further. The company's software can regulate the operation of equipment in the data center and improve energy efficiency. “This controls about 120 variables in the data center, including fans, refrigeration systems, and windows and other equipment,” said Hasabis.

He also said that this is only the beginning of the project. Currently, DeepMind knows that this method is effective and also begins to understand what information is missing from its artificial intelligence system. Therefore, DeepMind may ask Google to deploy new sensors in the data center to help the software get more information.