MOSCOW, January 20 Neural networks are better than weather forecasters at describing complex processes where there is no certainty, but artificial intelligence cannot create equations and check them for compliance with the fundamental laws of physics , said Roman Vilfand, scientific director of the Hydrometeorological Center of Russia.
“”There are several niches where a neural network or, better said, machine learning has a clear advantage: where there is uncertainty, where the laws are unknown – either they do not exist or they have not yet been formulated. For example, the description of subgrid processes, complex small processes. This is where the neural network is very good,” Vilfand said.
At the same time, the forecaster emphasized, artificial intelligence does not yet understand the basic laws of physics, so there should be no contradictions in the work of weather forecasters and neural networks, “synergy and interaction” are needed.
“The fact is that a neural network cannot conduct research in the same way as real meteorologists, mathematicians, and physicists do. Why? There are fundamental laws for the atmosphere: conservation of energy, conservation of mass, conservation of momentum. Equations, which scientists have developed correspond to these laws. The neural network cannot write out equations, partial differential equations of the second degree,” the meteorologist added.
Recent Comments