BUSINESS

Google Just Figured Out A Futuristic Way To Slash Its Energy Bill

After crushing human opponents in a board game, Google's DeepMind has a new project.

21/07/2016 4:12 AM AEST | Updated 21/07/2016 5:38 AM AEST
NEW! HIGHLIGHT AND SHARE
Highlight text to share via Facebook and Twitter
Connie Zhou/AP
It takes more than a mere human to make these glowing towers run on less power. 

Google is getting a lot smarter about how much electricity it uses.

The internet goliath is now using its DeepMind artificial intelligence technology ― the same one that recently trounced one of the world’s best human players of the notoriously difficult game “Go” ―  to cut electricity bills at its energy-guzzling data centers. Using DeepMind algorithms, researchers found more efficient ways to distribute electricity throughout Google’s vast network of servers, which make up about 5 percent of the cloud, the digital backbone of the internet.

The company wrote in a blog post Tuesday the machine learning helped it cut energy used for cooling by 40 percent, translating to a 15 percent decrease in overall power usage.

Artificial intelligence like DeepMind is essentially geared toward mastering a given task. It uses “machine learning” to recognize patterns in data and take action in response. 

DeepMind
This graph shows power usage effectiveness (PUE) at Google's data center. When DeepMind's machine learning (ML) is switched on, it plunges as the hardware uses less electricity to perform the same tasks. 

In the case of “Go,” a board game that’s more complex than chess, the DeepMind system processed and analyzed 30 million moves that had been played by experts in the past, allowing it to build a predictive model that could correctly anticipate the decisions a human player would make 57 percent of the time. DeepMind then played thousands of games against itself to develop its abilities ― and ultimately triumphed over the world’s top “Go” player.

But just because the system is good at a certain game doesn’t mean it can do anything else. The software lacks intuition, so it couldn’t apply its knowledge to chess, checkers or “World of Warcraft,” say. However, Google can apply the learning to other situations, like distributing electricity in its data centers.

The principle is basically the same: DeepMind is set to a task, and then it learns from a huge amount of data to make decisions and produce a desired outcome. In this case, that means finding ways to conserve power while keeping Google’s data centers live 24/7.

The announcement could create huge ripples throughout the web-hosting industry. Amazon, by far the biggest player with its Amazon Web Services, has a badly smudged environmental record, largely due to its slow adoption of renewable energy.

Statista
Amazon could learn a thing or two from Google. 

Google became the biggest corporate purchaser of green energy last year, and it’s investing aggressively to meet its goal of powering all its operations with renewables by 2025. The company built its servers with efficiency in mind, and has developed new ways to cool down the whirring-hot hardware with water instead of energy-thirsty air conditioners. That has helped Google get about 3.5 times more computing power out of the same amount of energy compared to five years ago.

Google said it aims to use the DeepMind technology to make its operations more efficient overall, as well as cut back on energy and water use at semiconductor factories.

“We are planning to roll out this system more broadly and will share how we did it,” the blog post says, “so that other data center and industrial system operators ― and ultimately the environment ― can benefit from this major step forward.”

Also on HuffPost
How Scientists Know Climate Change Is Happening

More On This Topic

Advertisement
Advertisement