Sunday, June 23, 2024

NewsHorizon

Where your horizon expands every day.

Science

Artificial Intelligence faces its own challenges when it comes to predicting the unpredictable. AI encounters difficulties in forecasting the unforeseeable, known as the Butterfly Effect.


A spectroradiometer image of a low-pressure system of clouds off the southeastern coast of Iceland. The clouds have a cotton-like appearance and form a loose counterclockwise spiral.

Source: Geophysical Research Letters

Traditionally, predicting the weather required using supercomputers to solve complicated math problems, which was time-consuming and energy-intensive. A newer approach involves using artificial intelligence (AI) to forecast future atmospheric changes. However, a recent study by Selz and Craig found that AI models do not take into account the butterfly effect, which limits the accuracy of weather prediction.

The concept of the butterfly effect explains how even small changes can lead to significant effects on the final outcome of a system. This can be seen in the analogy of a butterfly’s wings flapping in Brazil causing a tornado in Texas. In weather science, these rapid and unpredictable changes are often linked to convection and precipitation. However, in artificial intelligence systems, these initial variations may progress at a slower pace, making AI weather predictions less dependable.

Although AI may not be able to account for the butterfly effect in weather forecasting, it still has value in this field. This is because current atmospheric measurement errors are significant enough that the butterfly effect is not the limiting factor in predicting weather. In fact, AI is able to accurately model weather in midlatitude conditions despite these measurement errors, according to Selz and Craig. However, the accuracy of AI may vary in exceptional meteorological conditions. Scientists also have the potential to improve AI algorithms by creating additional training data that can teach them about the power of the butterfly effect. (Geophysical Research Letters, https://doi.org/10.1029/2023GL105747, 2023)

—Saima May Sidik (@saimamaysidik), Science Writer

Reference: Sidik, S.M. (2023). AI encounters its match: The butterfly effect. Eos, 104, https://doi.org/10.1029/2023EO230392. Retrieved from https://eos.org/articles/ai-meets-its-match-the-butterfly-effect-published-on-30-october-2023.

Text © 2023. AGU. CC BY-NC-ND 3.0

Images are protected by copyright unless otherwise specified. Unauthorized use without the explicit consent of the copyright holder is not allowed.