For years, Google has been creating AI fashions that enhance our understanding of the Earth. These fashions assist maintain Google companies updated. For instance, we analyze satellite tv for pc imagery to make sure map accuracy and supply search customers with up-to-the-minute alerts about climate and pure disasters.
As particular person fashions turn into extra highly effective, we discover that many real-world questions require a mix of domain-wide insights. Reply complicated questions resembling “The place is a hurricane prone to make landfall? The place are the areas most weak? How ought to I put together?” Requires inferences about photographs, populations, and environments.
Earlier this 12 months, we launched Google Earth AI to unravel this core problem. By combining our household of highly effective foundational fashions with a geospatial inference agent utilizing the newest Gemini mannequin, we will now carry out complicated real-world inference at planetary scale. This mannequin offers an in depth understanding of Earth primarily based on real-world information. Brokers act as clever orchestrators. Break down complicated questions into multi-step plans. Invoke these underlying fashions, question huge information shops, and use geospatial instruments to execute your plans. And eventually, combine the outcomes of every step to reach at an total reply.
At present we’re introducing a brand new innovation from Earth AI.
New picture and inhabitants basis fashions and technical particulars and analysis demonstrating state-of-the-art efficiency. Demonstration of a geospatial inference agent that makes use of these fashions to unravel complicated multi-step geospatial queries.
To study extra, learn the total technical doc Google Earth AI: Unlocking Geospatial Insights with Basis Fashions and Cross-Modal Reasoning. You too can categorical your curiosity and take part as we broaden entry to those new options for builders and companies.
