If there’s one thing that any artificial intelligence hates to handle, that would be a chaotic environment with numerous competing factors that also change rapidly. And if there’s one thing that encapsulates all of the above, that would be operating a flying air balloon.
There are many layers in Earth’s atmosphere, and the air is moving at different speeds and oftentimes in opposite directions. It’s the very reason why balloon pilots find it next to impossible to land in a specific place, as reading instruments and judging optimal actions takes time, and by then, things may have already changed.
Scientists are now developing a machine-learning system that can negotiate its movement in this dynamic environment, and stay within the range (50km) of a station. In this particular stage, the meteorology station would love to get its balloon back and use it again tomorrow. Or even better, keep it at the same altitude and position for extended periods of time, having it study small-scale effects, lighting, low-frequency sounds, etc.
As is the case with all neural networks of this kind, learning takes time, and learning in an environment that is underpinned by an extremely wide range of different conditions can take even more time. For this reason, the scientists are helping the unit by feeding it with weather forecasts, and also with noise data. This way, the actions are a bit smoother and allow the team to optimize carefully.
During the first tests, the balloons flew at heights of up to 20 kilometers, which was enough to test the directional changes of the wind and see how the trajectory shifts are handled. So far, the results have been quite exhilarating, demonstrating amazing possibilities in real-world applications. The team clarified that when the system goes to a different place than the one it was trained in, it almost definitely won’t perform as well immediately.
Flying a balloon is a feat for human pilots because there’s a lot of uncertainty stemming from the difficulty in the exact judgment of an ever-changing situation. This is one of the things that artificial intelligence can take from us without feeling sad about it.