The advent of advanced language models, such as OpenAI's GPT series and Google's LaMDA, paved the way for innovative and exciting human-AI collaboration opportunities. These models can generate human-like text, which can be harnessed to support various tasks ranging from language translation to question-answering. However, the extent of their capabilities for assisting humans in scientific discovery is not yet fully understood.

GPT, or Generative Pretrained Transformer, is a deep learning model developed by OpenAI. It is based on the Transformer architecture, a recurrent neural network that uses self-attention mechanisms to process sequential data. GPT uses a large corpus of unannotated text data to train a language model that can generate coherent and fluent sentences.

One of the critical advantages of GPT is its ability to handle long-range dependencies in natural language. This allows it to generate coherent and fluent text that is difficult for other models to produce. In addition, GPT has a large capacity for learning and can be fine-tuned for various natural language tasks, such as language translation, summarization, and question-answering.

But can this be used for scientific discovery? Turns out it can. An international team of scientists, including from the University of Cambridge, have launched a new research collaboration that will leverage the same technology behind ChatGPT to build an AI-powered tool for scientific discovery.

While ChatGPT deals in words and sentences, the team's AI will learn from numerical data and physics simulations from across scientific fields to aid scientists in modeling everything from supergiant stars to the Earth's climate.

Polythmathic AI

The team at Cambridge recently launched an initiative called Polymathic AI alongside the publication of a series of related scientific papers on the arXiv.org open-access repository.

Polymathic AI principal investigator Shirley Ho stated that this will completely change how people use AI and machine learning in science. According to Ho, the idea behind Polymathic AI is similar to how learning a new language is easier when you already know five languages.

The Cambridge report stated that starting with a large, pre-trained model, known as a foundation model, can be faster and more accurate than building a scientific model from scratch. That can be true even if the training data isn't relevant to the problem.

Siavash Golkar, a guest researcher at the Flatiron Institute's Center for Computational Astrophysics, remarked that Polymathic AI can show us commonalities and connections between different fields that might have been missed.

Advanced research

The Polymathic AI team includes physics, astrophysics, mathematics, artificial intelligence and neuroscience experts. Earlier, scientists used purpose-built AI tools and trained using relevant data. According to Francois Lanusse, a cosmologist at the Centre National de la Recherche Scientifique (CNRS) in France, despite rapid progress of machine learning in recent years in various scientific fields, in almost all cases, machine learning solutions are developed for specific use cases and trained on some very specific data.

The project will use data from diverse sources across physics and astrophysics (and eventually fields such as chemistry and genomics, its creators say) and apply that multidisciplinary savvy to various scientific problems.

Source: University of Cambridge

Want to publish your content?

Publish an article and share your insights to the world.

Get Published Icon
ALSO EXPLORE