Scientists have created a specialized AI assistant that will help in the study of Earth and other planets. It is capable of receiving data from geophysical instruments and using it to construct graphs and make assumptions. At the same time, requests can be made to it in normal human language.
AI for scientists
The new generation of AI has long sparked debate about whether it can really help people acquire new knowledge, or whether it is simply capable of stringing letters together to form words. But it seems that researchers at the University of Hawaii at Manoa have the answer to this question.
They created the Intelligent Data Exploring Assistant (IDEA) software platform, which combines the functionality of large language models such as ChatGPT with the capabilities of specialized neural networks designed to analyze large scientific data sets.Â
Based on IDEA, it is possible to create a variety of assistant instruments capable of understanding questions in plain language, analyzing queries based on data sets from various measuring devices, and generating graphs and assumptions that will provide scientifically agreed answers to scientists’ questions.
Earth and beyond
First and foremost, the platform is designed for studying Earth as a planet. Based on IDEA, scientists were able to create a prototype of the Station Explorer Assistant (SEA) and applied it to analyze data from tide gauges — instruments that measure sea level with high accuracy. Usually, special programs need to be written to process data from them, but the new AI has already demonstrated its ability to provide answers without this.
Researchers say that it is not perfect, still makes mistakes, and its work needs to be carefully monitored. But it still makes the work of scientists much easier. And this applies not only to oceanographic data.
Although SEA focuses on sea level data, the basic structure of IDEA is designed to work across a wide range of geological scientific disciplines. In one example from the study, researchers applied IDEA to atmospheric data from Mars — an area they had never worked with before — and were surprised at how easily the assistant adapted to the new dataset with only a change in instructions and data sources.
Although SEA is still a prototype, it is available online for scientists or university students to test and try out. Developers are encouraged to familiarize themselves with the IDEA framework on GitHub and experiment with adapting it to their own data or using other large language modeling services.
According to phys.org
