Candido and her team have been working on this problem for several years, since the company first launched as part of the Google X research lab in 2012. Loon is now a subsidiary of the parent company of Google Alphabet.
The big progress since then has been to apply reinforcement learning, something previously used in video games, to a real-world challenge, according to Marc Bellemare, main author of Nature article and a researcher at Google Canada. “Machine learning refers to the idea of taking data and making predictions about the outcome,” says Bellemare. “With reinforcement learning, we focus on the decision part. How can we increase or decrease based on this data? Not only [the AI controller] make decisions, but make decisions over time. “
Some experts believe that AI-powered balloons can also be used to monitor Earth’s environmental vital signs, such as checking melt into arctic permafrost, the exchange of greenhouse gases from tropical rainforests, or the atmospheric pressure and wind currents that give rise to powerful hurricanes in the Atlantic and the Pacific. The advantage of this kind of new AI-based navigation system is that balloons can be deployed from a distant launch pad – like Puerto Rico or Nevada, where Loon operates – and then actively ride the winds. to hit their target, much like a sailboat turns against the wind to cross the ocean.
“You can throw them where it’s convenient and cheaper, and then they could move around on their own,” says Scott Osprey, a climatologist at the University of Oxford, who was not involved in the research with Loon. Osprey sees a big role for stationary balloons in recording seismic waves from active volcanoes, for example, or even in interplanetary missions to explore cloudy atmospheres. “You can send a probe to Venus and look over orbiting cloud tops,” he says, “or park it there for months at a time and communicate with a satellite above.”
A trip to Venus can last a few years. Today, however, the new autonomous AI-powered balloons are stationed over Kenya, providing internet service to Kenya Telecom customers. The company recently broke a record for keeping one of its balloons at altitude for 312 days and is expanding its services to neighboring Mozambique in the coming months.
Google’s Bellemare says the advent of using reinforcement learning to navigate balloons for long periods of time will also open up all kinds of applications for scientific remote sensing and commercial projects. For him, this is another step in creating truly intelligent machines to perform difficult tasks without a human controller behind them. “What’s really exciting is the use of reinforcement learning,” he says. “It’s like trying to learn to ride a bike – it’s harder to write down the equations. It’s easier to try it. “
More WIRED stories