This news arrived at the heels of an environmental report indicating Google’s emission hikes by 48% due to machine learning.
One of the most sizeable technological break through in climate prediction
With the help of AI, it is possible to achieve new breakthroughs in the method of climate forecasting. Google Research today produces artificial intelligences: NeuralCGM that actually creates brand-new perceptions and assumptions about the alterations of our climate. This tool has developed a modelling technique grounded on the physics and AI of the system that can provide quick, efficient and accurate results concerning the Earth’s atmospheric condition.
Improved climate prediction accuracy
The older climate models assume a discretization of the entire globe and due to this assumption, they do not provide precise results of small scale conditions including cloud formation, turbulence and convection. It appears that the combination of physics-based simulations and AI does it. Based on the results obtained in 2020, it was established that NeuralCGM provided better annual temperature and humidity level forecasts with an accuracy ranging between 15% and 50% over the X-SHiELD model, which operates without an AI application. Even more strikingly, NeuralCGM came up with these predictions in 8 minutes while X-SHiELD came up with the same in 20 days.
Robust and machine-scalar AI model
Google said that NeuralCGM can work with a single TPU, whereas some high-resolution atmospheric models need a supercomputer, and thousands of CPUs. This means that “the model could be run on laptop and to the researchers whichever part of the world they are.” Greater efficiency could also lower the energy consumption by the model.
This comes a few weeks after, Google had revealed that the company’s carbon emissions for the past five years have been footed up by almost 50% by the intensive use of Artificial Intelligence. It is thus this innovation could be a solution to harmonize the advance in a technological world and the protection of the environment.