In an effort to cut energy use and improve performance, Google has decided to put an artificial intelligence system in charge of its data centre cooling after the system proved to be effective during its initial trial.
The search giant and its AI company DeepMind are now taking the project a step further. Previously recommendations were made by the AI and then implemented by human staff. Now though, the AI system will directly control the cooling in Google's data centres responsible for running services such as Gmail, YouTube and Google Search.
Since data centres use vast amounts of electricity to operate, even the smallest tweaks to areas like cooling can have a huge impact on cost savings. At the same time, Google's decision to use its DeepMind AI system to cool its data centres is also a great way for the company to promote its business.
The AI system works by using thousands of sensors to take a snapshot of the data cooling system. This data is then fed into deep neural networks which can predict how different choices will affect energy consumption in the future.
DeepMind then identifies the changes it could make to reduce energy consumption which are then sent back to the data centre, checked by the local control system and implemented to lower cooling costs.
Fortunately though, Google has put safeguards in place to ensure the AI is behaving correctly. Every time the AI comes up with a potential action, it is required to calculate how confident it is that the action is a good decision. Bad decisions are eliminated for consideration while good ones are then put into action.
Data Center Operator at Google, Dan Fuenffinger explained the company's reasoning behind letting an AI control how its data centres are cooled in a blog post (opens in new tab), saying:
“We wanted to achieve energy savings with less operator overhead. Automating the system enabled us to implement more granular actions at greater frequency, while making fewer mistakes.”
If Google's decision to let an AI control how its data centres proves to be successful in the long run, we will likely see other companies implement similar solutions in the future.
Image Credit: Scanrail1 / Shutterstock