- Researchers have developed a new AI tool to accelerate scientific discoveries
- LLM4SD explains the rationale behind its predictions for transparency
- Instead of replacing standard machine learning models, LLM4SD enhances them
An Australian research team led by Monash University has come up with a generative AI tool designed to accelerate scientific discoveries. Called LLM4SD (large language model 4 scientific discovery), retrieves the open source tool Information, analyzes the data and then generates hypotheses from them.
While LLMs are used in science, their role in scientific discovery remains largely unexplored, and unlike many validation tools, LLM4SD explains its rationale, making its predictions more transparent (and hopefully cutting down on hallucinations).
PhD. -Candidate Yizhen Zheng of Monash University’s Department of Data Science and AI explains: “Just as Chatgpt writes essays or solves mathematical problems, our LLM4SD tool sounds decades of scientific literature and analyzes laboratory data to predict how molecules behave -answering questions such as: or ‘Will this connection be dissolved in water?’ “
Simulation of researchers
LLM4SD was tested over 58 research tasks across physiology, physical chemistry, biophysics and quantum mechanics and exceeded leading scientific models, which improved accuracy by up to 48% to predict quantum properties that are essential for material design. Zheng said: “Apart from surpassing the current validation tools that act as a ‘black box’, this system may explain its analysis process, predictions and results using simple rules that can help researchers trust and act on its insight. ‘
PhD. -Candidate Jiaxin Ju of Griffith University said: “Instead of replacing traditional machine learning models, LLM4SD improves them by synthesizing knowledge and generating interpretable explanations”.
The team considers the tool as essentially “simulation of researchers”. Professor Geoff Webb from Monash University emphasized the importance of AI’s role in research. “We are already fully submerged at the age of generative AI, and we have to start exploiting this as much as possible to promote science while developing it ethically,” he said.
The research that was published in Nature machine intelligence and available to look at Arxiv Pre-Print ServerAt was a collaboration between Monash University’s Faculty of Information Technology, Monash Institute of Pharmaceutical Sciences and Griffith University.
