The article published in Advanced Intelligent Systems describes the study on a completely new tool for trend forecasting. The team of researchers used powerful language models (technology standing, among others, behind Chattempt) to map knowledge and direction of development in science. The results are promising.
The co -authors of the publication are, among others Maciej Tomczak and Stefanos Papanikolaou from the Nuclear Research Center (Nomaten/NCBJ), Dario Massa with Ideas-Ncbr and prof. Piotr Sankowski from the University of Warsaw. The team also included scientists from MIT and Korean Kist.
How does AI create a scientific map of the future?
The heart of the project was the analysis of over 10,000 specialized scientific articles in the field of nuclear materials, published over the two decades in the Journal of Nuclear Materials. Ai distinguished key concepts and a network of connections between them.
In this way, for each year, a separate one was created knowledge map. Each scientific term (e.g. “corrosion” or “titanium stop”) became a point on it, and the lines between them depicted how often these ideas appeared together in research. Analyzing how this map is evolving in time, scientists made two key discoveries.
Predictable rate of innovation
Researchers have noticed that the “distance” between the two concepts on the map – measured by the number of steps needed to combine one topic with the other – decreases in a predictable, exponential way.
In practice, this means that with the appropriate data You can estimate for how many years, today, distant research threads will become one hot, interdisciplinary trend. The model indicates future connections and development directions before they become obvious to observers.
Filter of groundbreaking ideas
The second pillar of the method is a kind of “news indicator”, which allows you to quickly assess the potential of each new publication. The system compares the number of new concepts brought by the publication on a map of knowledge with how loudly echoes (measured by the number of citations), and then classifies ideas in four categories:
- Innovative – A lot of new products and high citations. These are discoveries about a breakthrough potential.
- Controversial – A lot of new products, but initially a small interest. Risky ideas, but worth observing.
- Standard – Not much new, but high citations. A solid mainstream confirming known knowledge.
- Incremental – Little news and low citations. Minor improvements without revolutionary potential.
Innovation radar. Applications for business and science
The described method has great practical potential, especially for agencies granting grants and research and development departments, which would gain a tool for taking strategic decisions based on hard data. Since you can see which areas of the research will run in the coming years, priorities and design competitions and programs for future interdisciplinary trends can be determined in advance.
At the same time, thanks to the “news indicator”, the system allows wiser selection of projects – protects those risky, but promising (you can direct them to pilot paths) as well as objectively verifies the originality of ideasdistinguishing real innovations from duplication of known paths.
The authors honestly admit that their model is a prototype, proven for now in one specialized field. The next, natural step are experiments on wider data sets, e.g. from medicine, energy or research on artificial intelligence itself.
– This tool shows how we can try to anticipate research trends. For now, we have shown its action in a selected, quite specific area and confirmed that in a few cases it was able to predict which topics would interest researchers – says dr hab. Piotr Sankowski, director of the IDEAS Research Institute, research co -author.
– However, a real change would require its wider use on a larger number of data. If there is a demand, we will continue to work on it. Ultimately, the tool could be used to predict promising research directions and facilitate the assessment of whether we are dealing with a well -worn topic or something really breakthrough – he adds. If the model turns out to be universal, scientific institutions and research and development departments will gain a powerful tool for strategic navigation in the world of science and technology.
