In order to create an AI-powered instrument for scientific discovery, a group of scientists from across the world have joined forces to begin a new research project that will make use of the same technology as ChatGPT.
While ChatGPT works with words and sentences, the new project, known as Polymathic AI, will use physics simulations and numerical data from a variety of scientific domains to help scientists model anything from Earth’s climate to supergiant stars.
Shirley Ho, a group leader at the Flatiron Institute’s Center for Computational Astrophysics in New York City, US, and primary investigator for Polymathic AI, predicted that “this will completely change how people use AI and machine learning in science.”
Polymathic AI’s concept “is similar to how it’s easier to learn a new language when you already know five languages,” according to Ho.
It can be quicker and more accurate to start with a large, pre-trained model, referred to as a foundation model, rather than creating a scientific model from scratch.
That may hold true even in cases where there isn’t a clear connection between the training set and the current issue.
“We may have missed commonalities and connections between different fields, but polymathic AI can show us,” said co-investigator Siavash Golkar, a visiting scholar at the Center for Computational Astrophysics at the Flatiron Institute.
Experts in physics, astrophysics, mathematics, artificial intelligence, and neuroscience make up the Polymathic AI team.
The project called Polymathic AI will use data from various sources in the domains of physics and astrophysics (and eventually chemistry and genetics, according to its designers) to learn and will then apply this multidisciplinary expertise to a variety of scientific issues.
There are well-known restrictions on ChatGPT’s accuracy.
According to Ho, many of those problems will be avoided by Polymathic AI’s effort since it treats numbers as real numbers rather than merely characters on par with letters and punctuation. Real scientific datasets that capture the physics of the universe will also be used in the training data.
According to Ho, the project places a high priority on openness and transparency. Every detail is intended to be made public. In a few years, we hope to be able to provide the scientific community with a pre-trained model that can enhance scientific assessments in a wide range of issues and fields by democratizing AI for science.”
