X, formerly known as the Google X lab before spinning off into a new unit under parent Alphabet after a company restructuring, has tapped artificial intelligence to drive one of its most popular projects.
Project Loon, which looks to launch balloons into the stratosphere to provide internet access to users on Earth, made a major breakthrough recently. One of its balloons was able to stay up in Peruvian airspace for 98 days, an impressive feat considering the difficulty to keep a balloon at a certain spot for a long period of time.
The breakthrough was announced by the Project Loon team through its official Google+ page, where it stated that it was hard at work in the development of the navigation technology that its balloons will use.
The latest updates were put to the test this summer on one of Project Loon's flights, launching the balloon from a site in Puerto Rico and then having it travel to Peru. After 12 days of motion, the balloon was able to stay 20 kilometers (12.4 miles) over the city of Chimbote for 98 days.
The balloon was able to accomplish the feat through the use of artificial intelligence, specifically machine learning. The behavior of the balloon was altered to keep it in the air for a much longer period than was previously possible, with the balloon able to adapt to the wind conditions that would have sent it flying away into another spot in the past.
At the start of the project, the balloons were guided by handcrafted algorithms that programmed responses to variables such as location, wind speeds, time of day and altitude. The new algorithms under machine learning, meanwhile, allowed the balloon to analyze massive amounts of data and learn from them.
The predictions that the system will make are not perfect, largely because of the unpredictability of the weather in the stratosphere. To further improve the navigation of the balloons, X applied reinforcement learning, which makes the system continue collecting data on whether the balloon made the right decision after making the prediction.
The balloon that was able to stay up in Peruvian airspace, for example, was able to make just below 20,000 adjustments to its altitude over 98 days. Reinforcement learning, in a broad sense, is also what allowed Google's artificial intelligence program AlphaGo to beat South Korean world Go champion Lee Sedol, as the system analyzed millions of moves made by humans and then kept track of its right and wrong moves while practicing.
According to the official Google+ account of Project Loon, there is still a lot of testing in the future for its balloons. However, with the accomplishment of getting a balloon to stay at a precise location for 98 days, the optimism surrounding Project Loon is higher than ever.