Copyright © HT Digital Streams Limited All rights reserved. The Economist , The Economist 4 min read 04 Dec 2025, 15:36 IST ‘DeepSeek moment’ in AI vs humans: Artificial intelligence avatars crush human rivals in live sales (Pexels) Summary Delphi-2M can predict which of more than 1,000 of the following art conditions a person may face through MU detailed interrogation and physical investigation, which disease a given patient contracted. Much more difficult, but no less desirable, would be to identify which diseases a patient might develop in the future. That’s what the team behind a new artificial intelligence (AI) model, details of which were published in Nature on September 17, claim to do. Although the model, called Delphi-2M, is not yet ready for deployment in hospitals, its creators hope it could one day allow doctors to predict whether their patients are likely to develop one of more than 1,000 different conditions, including Alzheimer’s disease, cancer and heart attacks, all of which affect many millions each year. In addition to helping patients who are at high risk, it can also help health authorities allocate budgets for disease areas that may need extra funds in the future. The model was developed by teams at the European Molecular Biology Laboratory (EMBL) in Cambridge and the German Cancer Research Center in Heidelberg. It takes inspiration from large language models (LLMs)—such as GPT-5, which powers ChatGPT—that can produce fluent prose. LLMs are trained to detect patterns in enormous amounts of text scraped from the Internet, enabling them to pick the word most likely to come next in any given sentence. Delphi-2M’s creators reasoned that an AI model fed on large amounts of human health data could have similar predictive power. In many ways, the design of established LLMs was well suited to the task. One major adjustment needed, however, was to teach such a model to account for the time elapsed between events in a patient’s life. In written text, consecutive words follow each other immediately; the same is not true for diagnoses in a patient’s history. High blood pressure after a positive pregnancy test, for example, requires different interpretations depending on whether the two are separated by weeks—in which cases the pregnancy could be affected—or years. This adaptation was carried out by exchanging an LLM encoding a word’s position for one encoding a person’s age. (It was not without its mishaps: in an early version of the model, new diagnoses were sometimes predicted after a person had died.) Delphi-2M was then trained on data from 400,000 people from UK Biobank, a database containing arguably the world’s most complete human biological dataset. The model was given the timing and sequence of ICD-10 codes, the international medical shorthand doctors use to register officially recognized diagnoses, representing the 1,256 different diseases that appeared in the Biobank dataset. The model was then validated on data from the remaining 100,000 people in the Biobank before being further tested on Danish health records, which are known to be long-lasting and thorough. In this case, the team used data from 1.9 million Danes going back to 1978, ensuring a much more diverse and representative sample than the UK Biobank could provide. To judge the model’s performance, researchers measured its AUC (short for “area under the curve”, referring to a region in a probability map), in which a value of 1 would mean perfect predictions and 0.5 would be no better than random. For predictions of diagnoses within five years of a previous one, Delphi-2M performed on average at a value of 0.76 on UK data, with a small drop to 0.67 for the Danish data. Events that would often follow a specific previous one—death after sepsis, say—were more often correctly predicted, while more random external factors, such as picking up a virus, were harder to predict. Unsurprisingly, the model’s accuracy also dropped a bit over time: when forecasting ten years into the future, it averaged 0.7. Real-world applications remain far away for now. Delphi-2M will first have to go through a much more rigorous trial period that gives clinicians the opportunity to investigate whether it leads to better outcomes for their patients. That process can take many years. The Delphi-2M team is also working to update the model to allow it to take in more sophisticated data than chronological lists of diagnoses. As the UK Biobank also contains medical images and genome sequences, adding these data to the model can further improve its accuracy. As impressive as Delphi-2M appears, it’s not the only artificial health predictor in town. For example, an AI model called Foresight, originally developed at King’s College London in 2024, also uses patients’ medical history to predict future health events. (A larger version of the project was halted in June after concerns that NHS England did not seek the proper approvals when it gave the Foresight team access to the data.) The ETHOS model being developed at Harvard University also has similar aims. Although patients will have to wait to feel the direct benefits of Delphi-2M, even the preliminary version of the model already offers a potential treasure trove for biologists. His style of prediction reveals which conditions cluster together, which in turn can point to previously unexplored relationships between diseases. Future, stronger AI models may take that work even further. The possibilities are exciting, says Ewan Birney, a geneticist at EMBL. “I’m like a kid in a candy store.” Curious about the world? Join Simply Science, our weekly subscriber-only newsletter to enjoy our mind-bending science coverage. Get all the Business News, Market News, Breaking News Events and Latest News Updates on Live Mint. Download the Mint News app to get daily market updates. more topics #artificial intelligence Read next story
