AI may make fatal mistakes in weather forecasts

Scientists consider it one of the best AI developments that artificial intelligence can predict the weather for days in advance. In addition, it does all this faster and using less energy than traditional forecasting systems. However, a recent study pointed out that they are unable to indicate extreme storms for AI-based systems, which can have serious or fatal consequences.

• AI can accurately predict the typical weather conditions, quickly and energyfully.

• However

• A combination of traditional model and AI could be the key to future forecasts.

Researchers at the University of Chicago, the University of New York and the University of California Santa Cruz seeked the question of whether artificial intelligence can predict an extreme weather phenomenon that does not occur at all or rarely and not fed into its system.

According to the study Ai is doing a great job It is difficult in the typical daily forecast, but with rare or unprecedented events. These include, for example, in category 5 hurricanesoccurring once a century floods or heat waves that break historical records. In other words, it is the events that can cause huge disasters if those involved cannot prepare in time.

Why does the AI ​​encounter an obstacle?

THE ChatgptAi models like this are learned with huge amounts of data, and this is also done by the neural networks that focus on weather. Scientists feed them with decades of past weather observations and can predict what is expected soon. And with fresh data, very precise predictions are quickly prepared.

Pedram Hassanzadeh, an associate professor at Uchicago Geophysical Sciences, is considered one of the biggest achievements of AI-based weather models, but there is still room for improvement, because in the case of unexpected phenomena, predictions fail.

Can make fatal mistakes in artificial intelligence

This was tested by creating a artificial intelligence model with weather data that excluded hurricanes stronger than category 2. The model was then asked to advance a scenario that typically leads to a hurricane of 5. However, AI underestimated what is expected – writes Earth.

« You have always underestimated the event. The model knows something is coming, but it always predicts only a hurricane of 2 said Yongqiang Sun, a researcher at Uchicago, co -author of the study. Such a false estimate can also have fatal consequences.

What is the solution?

According to researchers, there is hope that once artificial intelligence can predict unexpected events. This is due to the fact that even if it has not happened in a given region, but globally, you can give better forecasts.

However, for the time being, experts say the combination of traditional and artificial intelligence is the next step.

How do you know more traditional forecasts? Traditional weather models use equations that describe the function of the atmosphere. They contain knowledge of physical and mathematical, as well as the interaction of heat, pressure and wind. In contrast, AI-based neural networks act as automatic filling devices, which are forecasts based on previously seen.