A cutting-edge large language model (LLM) outperformed human doctors in common clinical reasoning tasks including emergency room decisions, identifying likely diagnoses, and choosing next steps in management, according to a new study that used real emergency department data. The authors of the study – one of the largest studies to date to compare artificial intelligence and physicians on a wide array of clinical reasoning tasks – are clear that their results do not mean AI systems are ready to practice medicine on their own, or that doctors can be removed from the diagnostic process. The results do, however, raise urgent questions about the future evaluation and implementation of artificial intelligence (AI) tools in clinical care. For more than 65 years, difficult clinical diagnostic cases have been the gold standard for evaluating medical computing systems. Most recently, LLMs have surpassed earlier computational approaches on these complex cases. However, despite this progress, most medical studies of LLMs have examined narrow or highly controlled scenarios and often lacked direct comparison to the performance of human physicians in real-world clinical reasoning tasks. The rapid advancement of LLM-based medical tools now necessitates more rigorous evaluation.

 

Here, Peter Brodeur and colleagues comprehensively evaluated the diagnostic and treatment-planning abilities of an advanced LLM – the OpenAI o1 series – by comparing its performance to that of hundreds of physicians and earlier AI systems, across a range of clinical reasoning tasks. These included both standardized clinical cases and a real-world study involving randomly selected emergency room patients at a major emergency medical center in Massachusetts. Brodeur et al. found that, across all six experiments, the LLM model consistently matched or exceeded human performance in diagnostic and management reasoning. Notably, its advantage was most pronounced in early-stage emergency department triage, where clinicians must make rapid decisions with minimal information. While both humans and AI improved as more clinical data became available, the model demonstrated a distinct strength under conditions of uncertainty, using even fragmented, unstructured health record data effectively. According to the authors, LLMs are rapidly approaching, and in some areas surpassing, human-level clinical reasoning, and although AI-assisted decision-making is often viewed as risky, the findings suggest such tools – when used in collaboration with physicians’ assessments – could reduce diagnostic errors, delays, and disparities in access to care. However, the authors also note several important limitations of the study. For example, its focus was confined to text-based reasoning, whereas clinical practice depends heavily on visual and auditory cues, areas where current AI remains less capable. “Accuracy on a defined task is only one dimension of deployment readiness. Clinical AI must also deliver equitable, cost-effective, and safe outcomes, supported by accountability, transparency, and ongoing monitoring,” write Ashley Hopkins and Erik Cornelisse in a related Perspective. “Without robust demonstrated effectiveness, equity, and safety, many AI systems will remain insufficient for clinical use.”

 

***A related embargoed news briefing was held on Tuesday, 28 April. Recordings are available at https://aaas.zoom.us/rec/share/UpQfIbq6-1hq5AAk8728u9GxhtdmTQjaycbE58CtTdm0qFYI7xbABpyz9Lx7f073.O2z6hgNbQnfWC0rz  passcode: ytiHe.4M***

Targeting only a small group of well-connected individuals with smoking reduction interventions produces the largest reductions in smoking, researchers report. The findings provide actionable insights for designing social network-based interventions that leverage peer influence effectively. Peer influence plays a powerful role in shaping adolescent behavior, particularly in health-related habits such as smoking, as behaviors and norms spread through social networks during a developmentally sensitive period. Although prior research has shown that such influences can cascade through friendships and even friends-of-friends, how quickly the influence fades with social distance, and how social network structure affects that process, has remained unclear. Using Stochastic Actor-Oriented Models (SAOMS) and data from 3,154 students across two schools in the National Longitudinal Study of Adolescent to Adult Health (Add Health) dataset, Cheng Wang and colleagues investigated how peer influence decays across social distance. To better understand how behavioral effects spread and diminish, they evaluated the effects of different smoking reduction intervention strategies, targeting either random students or those most central within social networks. Wang et al. found that peer influence in the context of smoking reduction interventions extends up to three degrees of separation from intervention targets. Targeting 10 to 30% of highly connected individuals within a social network produced the greatest reductions in smoking, while increasing coverage beyond 40 to 50% yielded diminishing returns due to network saturation. Moreover, the authors discovered that social network structure mattered. For example, influence spread more widely and decayed more slowly in denser networks, while sparser networks showed more limited diffusion.

Researchers have uncovered a driver of methane emissions in livestock: a newly identified organelle, the hydrogenobody, which fuels methane production in the guts of livestock. The findings provide a cellular and molecular explanation for how single-celled organisms known as rumen ciliates, that live in the stomachs of animals like cows, contribute to methane emissions from these animals, offering a potential new target for tackling agricultural contributions to climate change. Methane is a highly potent greenhouse gas. A substantial fraction of human-caused methane emissions – particularly from ruminant livestock such as cattle and sheep – originates from microbial processes in the animals’ digestive systems. Within the rumen, a complex community of microbes supports digestion but also generates methane, with methanogenic archaea acting as the direct producers and rumen ciliates playing an important but still poorly understood role in amplifying these emissions. Despite extensive multi-omics research on the rumen microbiome, ciliates have remained understudied due to limited genomic resources, leaving key mechanisms unresolved.

 

To address this, Fei Xie and colleagues constructed a rumen ciliate genome (RCG) catalog comprising 450 genomes across multiple ruminant hosts and used it to analyze 1877 multi-omics datasets, alongside direct measurements from dairy cows. This large-scale integration linked ciliate abundance and activity to methane emissions and enabled validation in a real-world livestock setting. Xie et al. also discovered and experimentally confirmed a previously unknown organelle in rumen ciliates, the hydrogenobody (HB), which produces hydrogen while also regulating oxygen within the cell. By generating hydrogen and removing oxygen, the HB effectively supports methanogenic archaea while maintaining localized anaerobic conditions within the rumen. Moreover, its abundance varies with ciliate size and surface structure, indicating that different species occupy distinct ecological niches tied to micro-scale oxygen conditions. Notably, ciliates with higher HB abundance are associated with greater methane production, identifying them as potential targets for mitigation strategies aimed at reducing livestock emissions without broadly disrupting essential digestive functions.

Experts are warning that artificial intelligence (AI) must be carefully evaluated and governed before it is adopted widely in healthcare, saying rapid advances do not automatically translate into safe use for patients.

In a new study, researchers found a large language model (LLM) outperformed physicians across many common clinical reasoning tasks including emergency room decisions, identifying likely diagnoses, and choosing next steps in management.

MIT researchers found the expression of a gene can stimulate or suppress the activation of neighboring genes, by altering the biophysical properties of the DNA strand. The new findings could make it easier to control the output of synthetic gene circuits.