Scientists announced Wednesday the development of an AI system capable of predicting medical diagnoses years ahead of time, leveraging the same transformer technology that powers popular chatbots like ChatGPT.
The model, named Delphi-2M, analyzes a patient’s medical history to forecast the likelihood of more than 1,000 diseases well into the future, according to a study published in Nature by researchers from institutions in the UK, Denmark, Germany, and Switzerland.
Trained on data from the UK Biobank, which contains records of around half a million participants, the system uses neural networks originally designed for language tasks. “Understanding a sequence of medical diagnoses is a bit like learning grammar in a text,” explained Moritz Gerstung, AI expert at the German Cancer Research Center. “Delphi-2M learns the patterns in healthcare data—how diagnoses appear, in what combinations, and in what order—allowing it to make meaningful, health-relevant predictions.”
Charts presented by Gerstung showed the AI outperforming conventional risk factors, identifying individuals with significantly higher or lower chances of heart attack than age or lifestyle indicators alone would suggest.
To validate its performance, researchers tested Delphi-2M against data from nearly two million patients in Denmark’s national health database. Still, the team emphasized that the tool remains in the experimental stage and is not yet ready for clinical use. “This is still far from improved healthcare,” noted Peter Bannister, a health technology fellow at the UK’s Institution of Engineering and Technology, pointing to age, ethnicity, and outcome biases in the datasets.
Despite these limitations, researchers believe the system could one day transform preventive medicine. Gerstung suggested that such tools may guide early interventions and patient monitoring, while Tom Fitzgerald of the European Molecular Biology Laboratory said they could also help optimize resources in overburdened healthcare systems.
Current risk-assessment tools like the UK’s QRISK3 focus on specific conditions such as heart attack and stroke. Delphi-2M, however, has the potential to predict all diseases simultaneously and over extended timeframes, said co-author Ewan Birney.
Medical AI specialist Gustavo Sudre of King’s College London called the research “a significant step towards scalable, interpretable, and ethically responsible predictive modeling.” Interpretable AI remains a key goal in the field, as the inner workings of large-scale models often remain opaque even to their developers.
