Speakers: Dr Radmila Juric, Dr Eiman Almami, Dr Ibtesam Almami
In the last couple of years, knowledge management and discoveries in Biomedical Science have advanced through the application of Generative AI technologies across various domains, which range from drug discoveries, development and repurposing, patient care/healthcare delivery and pharmacovigilance to summarising biomedical research advances/results and creating collaborative transdisciplinary collaborations. However, there is overwhelming evidence that the precision of generative AI technologies in knowledge discovery/dissemination is debatable, in general. It does not look encouraging to claim that the GenAI technology is an answer to quick discoveries of or insights into knowledge, considering that data/knowledge is scattered across various sources with the complex semantic of data, its structures and their relationships. It is difficult to ultimately create or discover knowledge we need and trust, using GenAI and thus it is prudent to highlight that predictive inference might never be able to create/discover knowledge which is valid beyond reasonable doubt. There are numerus attempts to address this deficiency of predictive inference in GenAI, particularly if the technology relies on LLM and one of the latest ideas is to replace popular Retrieval Augmented Generation (RAG) with Knowledge Augmented Generation (KAG).
In this workshop we look at the opportunities offered by KAG in biomedical science from two different perspective. One is the natural extension of the GenAI technologies towards knowledge structures based on knowledge bases and graphs. The other approach is in defining GenAI models with extensions towards logic inference which offers both: structures not dissimilar to knowledge graphs and logic reasoning which could remedy the deficiency of predictive inference. We look at the challenges of using KAG and
(a) debate our software solution which accommodates GenAI augmentation with logic reasoning;
(b) outline which software platforms and GenAI models are suitable or needed to create operational environments in which results of applying GenAI are valid beyond reasonable doubt.
The field of biomedical science proved to be ideal to use in this workshop for a variety of reasons. There is an abundance of examples in biomedicine and life sciences to illustrate problems of using GenAI and justify debates in the workshop. However, the outcome is applicable to the use of GenAI technologies in general and across problem domains, particularly if our dependence in creating GenAI models will on LLM in future.
