British technology company DeepMind was acquired by Google in 2014, and the tech giant is already putting its new subsidiary to good use. New figures have revealed that Google has managed to reduce its enormous electricity bill by using DeepMind technology to make its energy use more efficient.
Google DeepMind, known as DeepMind Technologies before its acquisition, is an artificial intelligence company best known for creating an AI that can learn to play video games. The company hit the news in 2015 after its programme, AlphaGo, beat a human professional Go player for the first time. Unlike other games typically played by AI, Go is particularly hard for computers to learn because there are so many possible moves. Since then, Google has used this technology for a “game” of a different kind: reducing the company’s energy consumption.
Calculations suggest that Google’s total annual energy use equates to the average amount used by 366,903 family homes in the United States. A large amount of this energy is channelled to Google’s datacentres, which power all of Google’s services including mobile apps and cloud storage solutions. Google used the DeepMind technology to manipulate servers, cooling technology and power output in order to get the “highest score” – that is, the most efficient use of the energy. According to one of DeepMind’s founders, Demis Hassabis, the method of machine learning used in this operation was not much unlike the kind used by the AI to learn Atari video games.
The results of Google’s work with DeepMind are astounding: overall, Google has boosted its power usage efficiency (PUE) by around 15%. In some areas, the energy saving was absolutely staggering – in cooling, for example, Google reports that the AI consistently reduced energy output by 40%. The monetary value that comes with this saving has not been reported, but it is likely to be in the hundreds of millions of dollars.
This is great news for Google, who paid around $600,000 to purchase DeepMind (around £400,000) and can now use this saving to cover part of the cost. Environmental lobbyists will also be pleased with the news, as big corporations cutting their energy usage is a green move which is great for the health of the planet.
As for where Google will take DeepMind next, we aren’t quite sure. AlphaGo’s victory over European Go champion Fan Hui was one of the most notable moments for the company and in the entire history of machine learning. Fan Hui was a 2-dan professional at the ancient Chinese board game, out of a possible 9-dan, and the best player in Europe.
AlphaGo defeated Fan five games to none, on a full sized board without handicaps. As if this was not impressive enough, the programme then went on to defeat South Korean player Lee Sedol, widely regarded as the second in the world. Lee holds a 9-dan rank and was defeated 4 games to 1: Lee beat the programme in the fourth game but resigned the final game. This prompted the Korea Baduk Association, the organisation which oversees the game, to award the programme an honorary 9-dan grade.
It is possible that DeepMind may also be branching out of the video game sector and into health. It was reported earlier this year that the company had come into an agreement with Moorfield’s Eye Hospital to analyse eye scans of patients in order to find early signs of illnesses which can lead to blindness. This was somewhat controversial as it emerged that the data DeepMind could receive was highly sensitive, detailing a patient’s HIV status, mental health conditions and even if they had ever undergone an abortion. Since the public outcry, Google and the NHS agreed that the data should be anonymised and only the parts most relevant to the blindness research be accessed.
Blog courtesy of Syntax IT Support