A new study by researchers at Harvard Medical School and Beth Israel Deaconess Medical Center has found that advanced artificial intelligence systems can outperform human doctors in several medical reasoning tasks, including diagnosis and emergency care decisions.

The research compared physicians with large language models across a range of clinical scenarios. According to the findings, AI systems showed stronger performance in tasks such as identifying likely diagnoses, recommending treatment steps, and making decisions in emergency department settings where information is often limited.

Arjun Manrai, a co-senior author of the study, said the results demonstrate the rapid progress of AI in healthcare. He noted that the model surpassed both earlier systems and physician benchmarks in most tests. At the same time, he cautioned that better performance in controlled settings does not guarantee improved outcomes in real-world care.

The study evaluated OpenAI’s reasoning model, released in 2024, using a mix of published clinical cases and real-world emergency department data. Researchers presented the system with patient scenarios at different stages of care, from initial triage to later admission decisions. At each step, the AI was given only the information available at that point and asked to suggest diagnoses and next actions.

The results showed that the AI consistently outperformed doctors, especially in areas requiring structured reasoning and documentation. The largest gap appeared during the triage stage, when limited information makes decision-making more difficult. As additional data became available, both AI and physicians improved in accuracy, though the AI maintained an edge in many cases.

Peter Brodeur, a co-author of the study, said traditional testing methods such as multiple-choice questions are no longer sufficient to measure progress, as many AI models now achieve near-perfect scores. He added that newer evaluation approaches are needed to track further advances.

Despite the promising results, researchers stressed that the use of AI in healthcare must be approached carefully. They warned that while a model may correctly identify a diagnosis, it could also recommend unnecessary tests or interventions that might carry risks for patients.

The study’s authors called for further trials in real clinical environments to better understand how AI tools perform in practice. They also highlighted the need for investment in infrastructure and clear frameworks to support the safe use of such technologies.

The findings come with some limitations, as the analysis focused on a specific version of the AI model, which has since been updated. Researchers said additional studies are needed to compare different systems and explore how doctors and AI can work together effectively in patient care.

A new study has uncovered a key biological mechanism that may explain why coffee has long been linked to healthier ageing and a lower risk of chronic disease.

Researchers at Texas A&M College of Veterinary Medicine & Biomedical Sciences found that compounds in coffee interact with a protein in the body known as NR4A1, a receptor involved in regulating stress responses, inflammation and cellular repair. The findings shed new light on how coffee may help protect the body from age-related decline.

For years, studies have associated regular coffee consumption with a longer life and reduced risk of conditions such as heart disease, cancer and cognitive decline. Until now, however, the biological processes behind those benefits have remained largely unclear.

The research team identified NR4A1 as a critical target for several naturally occurring compounds in coffee, particularly polyphenols and other polyhydroxylated substances. These compounds bind to the receptor and appear to influence how it functions.

NR4A1 acts as what scientists call a nutrient sensor, responding to dietary compounds and helping the body adapt to stress and damage. It plays an important role in controlling inflammation, maintaining energy balance and promoting tissue repair — all essential processes in healthy ageing.

Stephen Safe, one of the study’s lead researchers, said the findings provide a clearer understanding of coffee’s protective effects. He explained that NR4A1 helps limit damage when tissues are under stress, and that its absence can worsen the effects of injury or disease.

Laboratory tests showed that coffee compounds reduced cellular damage and slowed the growth of cancer cells. When researchers removed NR4A1 from the cells, those benefits disappeared, strongly suggesting that the receptor is central to coffee’s protective action.

The study also highlights that coffee’s health effects are likely driven by more than caffeine alone. Decaffeinated coffee has also been linked to improvements in learning and memory, indicating that other components, including polyphenols, may play a significant role.

Recent research has suggested that moderate consumption of caffeinated coffee may also reduce anxiety, improve attention and vigilance, and lower levels of inflammation.

Scientists caution that while the findings are promising, more research is needed to determine how significant the NR4A1 pathway is in humans and how it interacts with other biological systems.

Still, the discovery offers an important step toward understanding why coffee remains one of the most widely studied beverages in nutrition science. It also reinforces the idea that compounds found in everyday foods and drinks can play a meaningful role in supporting long-term health and resilience as people age.