Successes and Failures of Artificial Intelligence Models in Biomedical Research

Researchers at the University of Virginia tested popular artificial intelligence tools such as ChatGPT, Gemini, and Claude on cell communication. The study revealed that artificial intelligence can find a large portion of known cell reactions, but struggles to predict responses to diseases or drugs.

Success Rate in Cell Communication

In a new study conducted at the University of Virginia, popular artificial intelligence tools were tested and asked to explain how cells in the body communicate. Focusing particularly on heart cells, the researchers determined that artificial intelligence was able to correctly identify up to 65 percent of known cell reactions.

Challenges in Drug and Disease Predictions

When it came to the stage of predicting how cells would react to diseases or new drugs, artificial intelligence models experienced major difficulties. It was observed that the prediction accuracy of artificial intelligence in this regard dropped as low as 6 percent, a situation that poses risks in terms of laboratory costs and patient safety.

Expert Opinions and Future Expectations

Dr. Jeff Saucerman from the UVA Department of Biomedical Engineering emphasized that technology is advancing, but knowing individual cell parts and combining them for future drug effects are different things. Stating that artificial intelligence models can be unreliable, Saucerman expressed that human scientists must have the final say.