More precise weather forecasts thanks to the new Microsoft-Ki

This is demonstrated by researchers with the participation of Johannes Brandstetter (University of Linz) with the Microsoft model “Aurora”. Appropriate: Today the book “What can artificial intelligence can do?” By Ai Pioneer Sepp Hochreiter.

As early as June 2024, Microsoft presented the “Aurora” model as an AI tool for the precise prediction of weather conditions and atmospheric processes. Now his skills have been presented in the renowned “Nature” Journal. The project showed that “human systems can be exceeded in predicting important weather phenomena,” said Brentstetter. With him, people could respond even better to extreme events in the future.

Ki -model indicated physics
As part of his earlier research activity in Amsterdam (the Netherlands), Brandster was a supplier of the project with a connection with the Microsoft Research Group “AI for Science”. At that time it was seen “that AI models such as Chatgpt became larger and more successful. At the same time you have not yet used your assets in the natural sciences”. But: “People prefer nothing to collect data,” says the physicist, for example in terms of the weather. So the possibilities of the large AI models were used “and they want to learn physics”.

“Aurora” is a “foundation model” for Earth Systems: “You have trained the model on general data, ie pumped quasi physical knowledge, and thus created a rough diamond,” the researcher explained from the Linz Institute for Machine Learning. The system was trained for more than a million hours with different geophysical data. In a second phase, the researchers times times their diamonds for different applications: “Aurora” was applied to various questions.

Five -day prediction of air quality and hurricanes
With the AI ​​tool, air quality can be predicted for every place in the world or the development of tropical vertebra storms and its potentially destructive paths. Ten-day predictions are possible for sea bonds or the weather in high resolution. The calculations need “less than a minute”: “In contrast to the conventional numeric systems, which have often been built with many people over the years, an AI model like Aurora can also be adapted to a certain application in just a few weeks,” brentter noticed.

It is not the first AI model for the weather forecast. In 2023, “Pangu-Weather” was already presented by the IT group Huawei “a breakthrough”, as there is in the newspaper, but with a relatively lower resolution. In a “nature” supporting video, Paris Perdialaris of the University of Pennsylvania (US) explains that “Aurora” is confronted with the basic challenge to develop very reliable, accurate and accessible prediction tools that need low computer -based sources. Only on the basis of historical data, “the model was able to predict all hurricanes more accurately in 2023 than all operational prediction centers,” said the main author. The high resolution of the AI ​​model for weather data also shows new potential, storms and extreme events very locally.

Great potential
In the next five to ten years it was the “Holy Grail” to build systems that work directly with the observation data supplied by external detection systems such as satellites and weather stations, ie the unprocessed data, and use them to get high resolution predictions for every desired location, according to Perdicaris.

Colleague Brandstetter also sees a great potential in Engineering to use the large AI models for data-driven prediction. So he founded the start-up “Emmi AI” in Linz in 2024. This is about the development of AI-based simulation models for industrial applications, for example for calculating wind tunnels or chemical processes.

Book about the benefits of the AI
“What can artificial intelligence do?” -A prominent Linz colleague from Brandster also follows this question in a new publication (ECOWING): The AI ​​pioneer Sepp Hochreiter. The developer of “LSTM Technology” (long-term memory), one of the basis for AI systems, follows the start of the research field and his own fascination. But also the role of AI in and for society. He does not classify AI-SCEPSIS and European challenges in this area to answer the question why we need the AI.

‘How can we cure cancer? How can we stop climate change? ” These are questions “on which we must concentrate”. For him, AI is not just a “helper” for more convenience, for writing e -mails etc. “No, AI has the potential to tackle the really big problems of humanity”, according to the author, she can help you to analyze very complex systems, to model and find “those intellectual skills”. The approximately 200 -page popular science book not only offers a classification, but is probably also an invitation for a common vision that could go with AI and people.