What Is DeepMind?
In 2014, Google acquired the artificial intelligence startup DeepMind, but it wasn't cheap. With a price tag of $500M, there must have been something special that Google saw in DeepMind that was worth acquiring. While the company hasn't produced any actual products for commercial use, they have focused on machine learning.
Machine learning is a type of artificial intelligence that aims to provide computers with the ability to learn new information without being directed to do so. Machine learning involves the development of computer programs that have the ability to teach themselves to grow and alter themselves when presented with new data.
DeepMind has experimented with progressing computers to think like humans do. They are known for their computer program that self-taught it to master Atari games (however, it couldn't beat humans in space invaders!).
Image courtesy of Google DeepMind
Simply put, DeepMind is programming computers to learn from visual data, just like humans do. You might be wondering, what good can come from this? Or, what benefits can be made from this advancement in computer science?
As of today, you might not visually see what machine learning is being used in, but it's all around you. Machine learning is used in things like identifying spam, stopping credit card fraud, image recognition, translation, and even predicting emergency room wait times.
Google's also using machine learning to help engineers find applicable solutions to tough issues. One of these issues is energy consumption.
Google's Current Carbon Situation
Google has a history of working to minimize their environmental impact. While a user uses Google's services (YouTube, Drive, Gmail, Search, Chrome, etc) for one month, they are using less energy than driving a car for one mile. (Isn't that crazy?)
Of course, Google came under fire in 2009 for the energy costs associated with each Google search and for the company's overall carbon footprint (a broad term indicating the amount of carbon that an entity is responsible for emitting). Still, the company claims that it's been carbon neutral since 2007, an achievement made possible through their purchase of carbon offsets (PDF). In Google's case, carbon offsets mean that they invest in energy research to help offset their carbon footprint.
In 2010, for example, Google invested $38.8 million in two wind farms on the Peace Garden wind farm in North Dakota, a that can produce 169.5 MW of power.
Image courtesy of Google
Another one of Google's large investments is in Germany. The Brandenburg-Briest Solar Park is one the largest solar facilities in Germany with a peak capacity of 18.65 MW. The facility can generate the same amount of energy required to power 5,000 homes.
DeepMind and Energy Efficiency
Demis Hassabis, the lead at DeepMind who leads around 200 computer scientists and neuroscientists at Google, stated that they are aiming for "solving intelligence, and then using that to solve everything else."
Guiding this credo, the project has been able to help save 40% of their energy consumption used to cool its data centers.
Since its acquisition, DeepMind has been used for a few research and AI projects, but recently it has been put in charge of taking on a different task pertaining to Google's data centers that will make them more efficient. The current goal is to build a new framework around Google's data centers that will read a tremendous number of variables (about 120, to be exact), analyze them, and then optimize efficiency based on the analysis.
Image courtesy of Google
This AI has worked through every possible solution in terms of finding the most efficient method of cooling the data centers through analyzing sensor data in the server racks such as average temperatures, up and down times, and pump speeds. The next steps for the engineers working on this project are to identify more locations where data can be extracted from and analyzed to calculate an even better efficiency.
Google stated that, after considering inefficiencies not related to cooling issues, the 40% reduction in energy usage converts to about a 15% decrease in their overall power usage efficiencies (PUE). Google consumed a whopping 4,402,836 MWh of electricity in 2014—with DeepMind's 15% reduction, hundreds of millions of dollars can be saved and the size of their carbon footprint can be reduced.
Applications of DeepMind's Artificial Intelligence Algorithm
Knowing Google, they won't just end their DeepMind project with cooling their data centers to improve their efficiencies and reduce energy consumption. In fact, DeepMind has stated as much on their blog.
There are many applications DeepMind could be utilized in. In the future, we could see machine learning help improve power plant conversion efficiency, reduce semiconductor manufacturing energy and water usage, or even increase the efficiencies of photovoltaics.