Connecting artificial intelligence with the tools of astrophysical research

Investigating a scientific question is often a complex process. As an astrophysicist studying some of the most energetic phenomena in the universe, I know that finding answers requires much more than reading scientific papers or understanding existing theories. Astrophysical research involves collecting observations, analyzing data, building physical models and comparing their predictions with observations. These tasks require different scientific tools, databases and modeling software and can take considerable time even for experienced researchers.

This challenge motivated the development of AstroGenesis, a multi-agent AI system designed to connect different stages of astrophysical research within a single environment. Our goal is to make scientific investigations faster and more accessible. A paper describing the work is posted to the arXiv preprint server.

Why scientific research needs more than answers

Over the last few years, large language models (LLMs) have changed the way we interact with information and technology. They can explain complicated scientific processes, summarize research papers and help researchers write code. These capabilities are already useful in science.

However, there is an important difference between explaining how scientific research is performed and actually carrying out the necessary calculations and analyses.

If we ask an AI system to investigate the radiation produced by a distant black hole, it can provide a detailed explanation of the physical processes involved, describe previous studies and suggest possible interpretations. But a scientific investigation requires additional steps. We need actual observational measurements, information about their uncertainties and physical models that can be tested against the data.

A convincing and detailed explanation is not necessarily a scientifically verified result.

For me, this distinction raises an important question: Can we connect the capabilities of modern AI with the scientific tools and methods needed to carry out actual research?

From an idea to AstroGenesis

To explore this possibility, I worked with an international team of astrophysicists and AI researchers to develop AstroGenesis. Our central idea was to connect modern AI with the scientific tools and methods that astrophysicists use in their daily research.

Rather than simply generating answers based on existing knowledge, AstroGenesis is designed to perform scientific tasks by accessing research publications, retrieving and analyzing observational data, and carrying out calculations using physical models. Researchers can interact with the system through natural language, asking scientific questions much as they would ask a colleague. AstroGenesis then identifies the tasks needed to address the question and coordinates specialized AI agents to perform them using the appropriate scientific tools.

Instead of relying on a single AI model to perform every task, AstroGenesis combines several specialized agents. A central supervisor coordinates their activities and assigns tasks to the appropriate agents. One agent searches scientific literature and identifies relevant studies. Another retrieves astronomical observations and performs data analysis. A modeling agent calculates predictions from physical models and compares them with observations, while another agent helps researchers explore possible scientific interpretations and research questions.

Making complex physical calculations faster

One of the most computationally demanding parts of astrophysical research is theoretical modeling. To understand astronomical observations, we often use physical models that involve solving complex equations. These models are powerful tools for investigating astrophysical processes and can provide detailed and accurate predictions. However, the calculations are often time-consuming.

In AstroGenesis, we use an innovative approach to address this challenge by integrating pretrained neural networks into the physical modeling process. We train neural networks using results produced by numerical physical models. Once trained, these networks can rapidly approximate the predictions of the original models without solving the full set of equations each time. This does not mean replacing physics with artificial intelligence. The underlying physical models remain essential, while neural networks provide fast approximations of their predictions.

This approach can significantly reduce the time needed for theoretical modeling while maintaining good accuracy within the parameter ranges where the networks have been trained and validated. More importantly, it allows researchers to explore a much larger range of model parameters in a short time. This can help identify possible physical solutions that might have been missed when using slower numerical calculations over a more limited parameter space.

From observations to physical interpretation

Consider a researcher studying the radiation emitted by an astrophysical source. Traditionally, this involves searching the literature, collecting and analyzing observations from different archives, and running numerical models to investigate the physical conditions responsible for the observed radiation.

With AstroGenesis, the researcher can describe the scientific question in natural language. The system then coordinates its specialized agents to retrieve relevant publications, analyze observational data and calculate predictions from physical models.

The goal is not simply to generate an answer but to connect it to actual observations and physical calculations that researchers can examine.

Another important goal is to democratize access to advanced scientific research tools. By reducing the technical barriers involved in data analysis and physical modeling, we aim to make these capabilities more accessible to students and researchers, including those at smaller institutions.

Toward a different way of doing science

I believe the most interesting role of artificial intelligence in science is not simply to provide faster explanations of what we already know. Its greater potential is helping researchers work more efficiently with the enormous amount of scientific knowledge, observational data and computational tools available today. By connecting these resources, AI can help researchers explore scientific questions, test ideas more quickly and investigate problems that would otherwise require considerable time and effort.

There are still important challenges. AI-generated interpretations must be checked, observational uncertainties must be respected, and approximate models must be validated against the physical calculations they represent. Scientific judgment remains essential.

Our aim is not to replace researchers but to give them tools that make sophisticated scientific investigations more accessible and allow more time for the questions that matter most: What do the observations tell us, which physical explanations are possible and what should we investigate next?

Ultimately, I see this as a step toward a different way of doing science, where artificial intelligence becomes an active part of the research process rather than simply a tool for providing information.

This story is part of Science X Dialog, where researchers can report findings from their published research articles. Visit this page for information about Science X Dialog and how to participate.

Who’s behind this story?

Lisa Lock

BA art history, MA material culture. Former museum editor, paramedic, and transplant coordinator. Editing for Science X since 2021.

Full profile →

Robert Egan

Bachelor’s in mathematical biology, Master’s in creative writing. Well-traveled with unique perspectives on science and language.

Full profile →

Narek Sahakyan is an astrophysicist and Director of ICRANet-Armenia. His research focuses on high-energy astrophysics, relativistic jets, the physical mechanisms responsible for radiation from extreme cosmic sources, and the application of artificial intelligence in scientific research. He leads the development of AstroGenesis, an international research initiative exploring how artificial intelligence can be integrated with scientific literature, observational data, and physics-based modeling to support astrophysical investigations.