 |
Center for Nonlinear Studies |
The scientific literature is the key mechanism by which scientists communicate new knowledge. Due to the volume of output from the scientific community – millions of papers each year – it can be very challenging to find, use, and synthesise key insights from this vast resource. Tasks ranging from systematic reviews to protein function prediction to hypothesis generation can benefit from organisation, mining, and modelling of scientific information expressed in text using computational approaches. I will describe how artificial intelligence (AI) and natural language processing (NLP) methods can be used in positive ways to structure and analyse key information described in textual resources, with reference to a range of biomedical and biochemical application contexts. On the flip side, the advent of generative AI systems including generative large language models (LLMs) in late 2022 has resulted in rapid changes to our scientific ecosystem. With LLMs that have ingested and been trained on arguably most published scientific knowledge, they are increasingly being seen as replacements for traditional research strategies such as librarian-supported information management or search engines, and to summarise bodies of evidence. Furthermore, they are being used for other parts of the scientific process, including ideation, design of experiments, data analysis and interpretation, writing of new papers, and even peer review. This raises a number of concerns about the integrity of science, and the implications for our roles as scientists. This talk will present and explore this complex duality between the promises and the perils of AI for science.
Host: Anna Llobet (allobet@lanl.gov) and Sean Kuvin (kuvin@lanl.gov)