Google deploys DeepMind AI to cut data centre power consumption

Machine learning technology improves power efficiency by a claimed 15 per cent

Google has deployed machine learning know-how picked up from its DeepMind acquisition to improve data centre efficiency and reduce power consumption and therefore costs.

The artificial intelligence (AI) technology is applied to the management of the servers and other electronics at the company's proprietary data centres, and is used specifically to fine-tune the cooling systems.

"It controls about 120 variables in the data centres. The fans and the cooling systems and so on, and windows and other things," DeepMind co-founder Demis Hassabis told Bloomberg at a recent conference on AI in New York.

He added that the company's technology was deployed in Google data centres only in recent months, but has already resulted in a 15 per cent improvement in power use efficiency.

The software uses an adapted form of the machine learning software developed by DeepMind to play Atari 2600 video games. Deployed in the data centre, the software learns how the servers work and manipulates fans and other elements to reduce their use when not needed.

Hassabis did not go into greater detail about the deployment or say whether Google will make it available to other companies. However, the improvements in efficiency will go a long way to paying for the DeepMind acquisition, given Google's estimated four gigawatts of annual electricity consumption.

Google paid around $400m to acquire the London-based AI startup in January 2014. It was founded by Hassabis in September 2010, along with Mustafa Suleyman and Shane Legg.

Hassabis and Legg were machine learning experts, while Suleyman was an entrepreneur who handled the business side of the company.

However, the use of DeepMind's technology has been controversial, not least because it was deployed by Google to analyse some 1.6 million NHS patients' records in a deal with the Royal Free NHS Trust revealed earlier this year.