Published: 28-09-2026 11:58 | Updated: 28-09-2026 12:39

Award highlights AI framework for safer healthcare

Image:

Caption:

Farhad Abtahi, receiving the award from Joao Monteiro, Chief Editor of Nature Medicine and Alexandre Loupy, Director of the Paris Institute for Transplantation and Organ Regeneration, and Conference Chair.

An abstract by researchers at Karolinska Institutet has received the Best Abstract Award at the Nature Conference "Redefining Healthcare in the Age of AI" in Paris. The work describes a framework called MEDLEY, which uses differences between AI models to make uncertainty in healthcare decisions more visible.

Researchers increasingly explore how large language models can support healthcare, but ensuring that such systems are reliable remains a challenge. At the Nature conference Redefining Healthcare in the Age of AI, held in Paris, France, on 8–10 September 2026, Farhad Abtahi, Senior Research Infrastructure Specialist at the Department of Clinical Science, Intervention and Technology (CLINTEC) received the Best Abstract Award for “Bias as a feature, not a bug: leveraging multi-model disagreement for trustworthy clinical AI”. The abstract was also selected for an oral presentation.

The presentation introduced MEDLEY, a framework that treats disagreement between different AI models as potentially useful information rather than something that should be eliminated.

According to the researchers, systematic differences between models may reveal weaknesses in reasoning, gaps in training data, temporal or population-specific biases, and situations that require further validation. Such information could support the development of AI systems intended for clinical use.

During deployment, the framework can present clinicians with a consensus view, plausible alternative responses, minority opinions and information about the origin of each output. This could help users identify uncertainty, verify important recommendations and recognise cases that require additional review.

The researchers emphasise that MEDLEY is intended to complement, not replace, clinical expertise. The framework would need to be used together with clinical validation, monitoring of performance in different patient groups, human-factors testing and clear accountability structures.

A scientific article describing MEDLEY has been published in Frontiers in Artificial Intelligence, and an online demonstration of the framework is available. The research team welcomes collaborations aimed at validating the framework in healthcare settings.