Google deploys AI to autonomously optimize data center cooling systems, improving energy efficiency.

Google has given autonomous control of cooling systems at several of its data centers to an artificial intelligence algorithm, marking what Mustafa Suleyman, head of applied AI at DeepMind, calls "the first time that an autonomous industrial control system will be deployed at this scale, to the best of our knowledge." The algorithm was initially deployed in a recommendation capacity, where it suggested adjustments to fans, ventilation, and other cooling equipment to lower power consumption. During this phase, the system delivered energy savings of around 40 percent in those cooling systems, with human data center managers retaining the decision to implement or reject each recommendation.

The AI system uses reinforcement learning—a technique in which algorithms learn through trial and error—the same approach that produced AlphaGo, DeepMind's Go-playing program. DeepMind trained the algorithm using information gathered from Google data centers, allowing it to determine what cooling configurations would reduce energy consumption. According to Joe Kava, vice president of data centers for Google, the project could generate millions of dollars in energy savings and may help the company lower its carbon emissions.

Although the algorithm now operates independently, human oversight remains built into the system. Data center managers can observe the algorithm's actions, check its confidence level about the changes it proposes, and intervene if necessary. Safety controls prevent the system from making adjustments that would adversely affect cooling. Kava noted that managers trusted the earlier recommendation system and had few concerns about delegating greater control to AI.

Data center energy consumption has become a critical concern for the technology industry. A 2016 report from the US Department of Energy's Lawrence Berkeley National Laboratory found that US data centers consumed about 70 billion kilowatt-hours in 2014—approximately 1.8 percent of total national electricity use. The same report projected that while efficiency gains are nearly offsetting the energy consumption of new data centers, total use would reach around 73 billion kilowatt-hours by 2020.

However, Jonathan Koomey, a leading expert on data center energy usage, cautioned that cooling accounts for only about 10 percent of a data center's energy consumption. He suggested that applying machine learning to optimize the computing loads from power-hungry computer chips inside data centers could prove far more significant. "The possibilities on the compute side are tenfold bigger than for cooling," Koomey said, expressing eagerness to see Google and other major technology companies apply such optimization tools to their computing workloads.