In the age of Artificial Intelligence (AI), the scientific community is questioning the real potential of advanced AI language models in medical practice. Recent research published in the British Journal of Haematology, conducted by the University of Milano-Bicocca in collaboration with the bone marrow transplant centres of IRCCS San Gerardo dei Tintori and Niguarda Hospital, has shed light on this challenge.
The study, conducted by Dr. Ivan Civettini (University of Milano-Bicocca) and Dr. Fabrizio Cavalca (IRCCS San Gerardo Foundation) under the supervision of Prof. Daniele Ramazzotti (University of Milano-Bicocca) and Prof. Gambacorti-Passerini, Professor of Haematology, (University of Milano-Bicocca), evaluated the effectiveness of advanced language models, such as GPT, Palm2 and LLama, in the delicate decision-making phase of haematopoietic stem cell transplantation. Using complex haematology clinical scenarios, the AI was subjected to rigorous analysis and its responses were compared with those of experienced transplant physicians and haematology residents.
The results of this research demonstrate the potential of artificial intelligence in the context of stem cell transplantation, with excellent responses in transplant eligibility and donor selection. However, there are significant limitations in treatment planning and risk assessment. Human clinical experience remains unsurpassed in this context, but the study predicts that with future refinements, these language models could become a valuable tool for doctors in their daily practice.