The danger of avalanches can be assessed with artificial intelligence just as well as with humans, mistakes included. This is what the Swiss Institute for Snow and Avalanche Research (SLF) in Davos reports after a three-year research project. The good news: machine-trained algorithms and people have different strengths and weaknesses and can therefore complement each other well.

The SLF has been working with such models for three years. The model's predictions are often good. “Sometimes they are clearly not, but we are also wrong sometimes,” said avalanche warning Frank Techel. Algorithms have been used for a long time, but what is new is that they analyze and evaluate the results of many models and provide their own assessment.

Humans would have the advantage that, unlike algorithms, they could, in addition to the data and models, also take current observations and feedback from people in the field into account when assessing the situation. On the other hand, due to time constraints, humans could only include the most relevant data in their analysis, while computers take all information into account. “The good thing is: the models make different mistakes than we do,” said Techel.