Taking into account inefficiencies in the hardware, the AI managed to save 15 percent of Google's total power usage.
One of the biggest problems people have found in Pokemon Go is the fact that services are unstable. This is thanks to developer Niantic underestimating the app’s popularity,and consequently dedicating fewer servers than required. Why didn’t they just take the safe road and leave a huge margin for error with extra servers? Well, that’s because servers are notorious for guzzling energy like a penguin at a sushi restaurant.
That’s why, Google is experimenting on trying to cut down on the cooling bill from it’s huge data centers. The twist here, is that they’re using the DeepMind AI to do it.
According to Bloomberg, Google put its in-house AI to work on managing power usage in some parts of its data centers. The result was a 40 percent reduction in power usage for cooling, something the company described as a “phenomenal step forward." Even after taking into account things like electrical losses and “non-cooling inefficiencies” the overall amount of power saved was a very respectable 15 percent.
That number may not seem high enough to make a huge difference, but it’s important to keep in mind the scale of Google’s servers. Specifically, Google used over 4.4 million MWh of power in 2014, saving just 15 percent of which could result in saving millions within a few years. In fact, with that sort of steady saving, Google may soon even overwrite the US $600 million it paid for DeepMind in 2014.
According to Bloomberg, DeepMind co-founder Demis Hassabis said DeepMind controlled about 120 points in the data center as part of the experiment, including fans, cooling systems, windows, and others. The neural network then analyzed the efficiency of each method of cooling in an area, through sensors placed among the server racks, and took steps to make changes where it deemed necessary. DeepMind’s engineers believe the AI can do even better, and any current inefficiencies stem from their own, with regards to where sensors were placed or more might be needed.
If the long term experiment continues to be a success, Google says it’ll take DeepMind even further. In a blog post for the experiment engineers said, "Because the algorithm is a general-purpose framework to understand complex dynamics, we plan to apply this to other challenges in the data centre environment and beyond in the coming months. Possible applications of this technology include improving power plant conversion efficiency (getting more energy from the same unit of input), reducing semiconductor manufacturing energy and water usage, or helping manufacturing facilities increase throughput."
Of course, the one other thing that you should maybe also take away from this is that we continue to build up AI’s idea that they’re better than humans at every job. You know what happens when machines in charge of basic features of our life gain sentience? Very very bad things. Let’s hope that AI kill switch comes around before this gets too far.
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
 "artificial intelligence (AI), DeepMind, Google, Data Center, server, Power Usage Effectiveness")
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
)
){kind=small}
){kind=small}
)
)
)
)
)
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}