AI Spots Novelty: Jülich Team Wins Global Contest

8 June 2026

A research team from Forschungszentrum Jülich has won the international "Metascience Novelty Indicators Challenge". The scientists at Jülich Systems Analysis developed a method that enables artificial intelligence to assess the novelty of scientific publications - in other words, the extent to which a study advances scientific knowledge. For their success, the team has been awarded prize money of £300,000 to further develop the method.

The challenge was organised by the UK Metascience Unit (UKRI) in collaboration with international partners. The aim was to develop a scalable method for assessing the novelty of research articles at the time of their publication.

To this end, the organisers provided a dataset of 100,000 recent scientific publications. Experts from the respective disciplines assessed their novelty independently of one another. The task for the participating teams was to predict these expert judgements as accurately as possible - without knowing the assessments.

The Jülich approach achieved the best results across all evaluation criteria.

"Until now, the ability to assess what is truly novel and valuable in a scientific paper has been limited to human experts," says Dr.-Ing. Jann Michael Weinand, Head of the Integrated Scenarios Department at the Institute of Climate and Energy Systems - Jülich System Analysis (ICE-2). "Our approach shows that modern AI systems can support this task with astonishing reliability."

AI analyses content rather than citation counts

Unlike many established research metrics, the Jülich system does not assess how often a paper is cited later on. "Metadata is not sufficient to assess novelty at the time of publication. Our system therefore examines the content of a study and relates it to the state of knowledge at the time of its publication," says project leader Jan Göpfert, also from ICE-2, who developed the approach together with his colleague Samuel Kieling.

To do this, the system first analyses the study itself as well as selected scientific papers to which it refers. On this basis, the AI reconstructs the state of knowledge at the time of publication, including known gaps in research. It then assesses the contribution made by the new study. Does it introduce a new method? Does it deliver surprising results? Does it solve a previously unsolved problem? In doing so, the system deliberately collects arguments both for and against the novelty of a paper and weighs them against one another.

In the end, the AI assigns a novelty score between 0 and 100. It also provides an interval indicating how confident the model is in its assessment. A detailed written justification then makes the evaluation transparent. "The biggest challenge was defining novelty in a meaningful way. For us, novelty does not simply mean dissimilarity. What matters is a work's contribution to scientific progress," says Kieling.

Earlier visibility for important research

The number of scientific publications continues to grow rapidly. At the same time, an increasing number of papers are being produced with the help of AI tools. This makes it increasingly difficult for researchers, academic journals and funding organisations to identify particularly relevant contributions at an early stage.

This is where the novelty indicator could help in future. Research with particularly high potential for generating new insights could be identified during the peer-review or publication process - rather than only years later, when its significance becomes apparent through citation metrics.

"We hope this will particularly benefit research that is often overlooked by traditional metrics," says Kieling. "Our goal is not to replace human judgement. Rather, AI should help draw attention to potentially important research and support better-informed decisions."

Furthermore, the Novelty Indicator opens new possibilities for metascience, i.e. the scientific study of the research system itself.

Prize money enables further development

With the prize money of £300,000, the team intends to further develop the existing prototype into a reliable scientific tool. The novelty indicator is intended to be transparent, resistant to manipulation, and must not exacerbate existing inequalities within the scientific system.

In the long term, the researchers envision applications far beyond scientific publications - for example, in the context of patents or the identification of new and promising research questions and hypotheses. "At the same time, this development raises fundamental questions: What role should AI play in scientific decision-making in the future? And how can we ensure that scientific evaluation and progress remain transparent and traceable?" says Göpfert.

The work of the Jülich researchers demonstrates that AI is now capable of far more than analysing data or summarising texts. It is increasingly able to evaluate scientific research itself - opening up new possibilities for the science of tomorrow.

About the Metascience Novelty Indicators Challenge

The Metascience Novelty Indicators Challenge was hosted by the UK Metascience Unit at UK Research and Innovation (UKRI) and Coefficient Giving. Partners in the competition were the Science Policy Research Unit (SPRU) at the University of Sussex, the research and consultancy institute RAND Europe, and Challenge Works, global leaders in the design and delivery of challenge prizes, and part of the research and innovation foundation Nesta. £300,000 prize was provided by Coefficient Giving.