Media contact: Dan Watson
(text and background only visible when logged in)
As strange as it sounds, the key to understanding life’s origins might lie in artificial intelligence. At least, according to a new approached being pursued by researchers at Georgia Tech.
School of Electrical and Computer Engineering (ECE) Assistant Professor Amirali Aghazadeh and Ph.D. student Daniel Saeedi have developed AstroAgents, an AI system that analyzes mass spectrometry data — detailed chemical compositions from meteorites and Earth soil samples — to generate novel hypotheses about the origins of life on the planet.
What sets AstroAgents apart is its use of agentic AI. Unlike traditional AI systems that perform fixed tasks, this agentic system is designed to pursue a scientific goal. It draws from astrobiology literature, interprets complex data, and proposes original ideas that researchers can investigate further.
Their paper, recently featured in the journal "Nature", is opening new possibilities for how scientists explore questions that have remained unanswered for decades.
In the Q&A below, Aghazadeh and Saeedi explain how AstroAgents analyzes space chemistry, what it’s revealing about the possible origins of life on Earth, and what they hope to explore next.
(text and background only visible when logged in)
(text and background only visible when logged in)
What exactly is AstroAgents?
Amirali Aghazadeh (AA): Put simply, it’s an AI system that generates hypotheses about how life started on the Earth. There is this really big question that's been out there, which is, what are the building blocks of life? We still don't know on the molecular level how life started. In this project, we are interested in developing “signatures” for life. When we explore other planets, such as Mars and Jupiter, we want to have a “life detector” device on spacecraft that can use the signatures informed by this research to find evidence of life.
So, how does it work?
AA: We developed an agentic AI system to analyze mass spectrometry data, which is basically detailed chemical data. In this case, we were analyzing data of meteorites and soil from Earth. The AI system itself is trained on the data from astrobiology literature.
There are so many organic materials in the data from these samples, you can think of it as a soup of organic materials. The AI’s job is to look for similarities and differences in this data and use the literature to come up with novel hypotheses.
That’s the “agentic AI” part, it means the systems doesn’t just follow instructions but actively pursues a goal. The goal here is to generate new, scientifically plausible hypotheses about the origins of life through the data.
Daniel Saeedi (DS): The system is built on two pretrained large language models, Claude 3.5 from Anthropic and Gemini 2.5 from Flash. We used their application programming interface to power our system.
How did this project come about?
AA: We were working on developing machine learning (ML) models to identify new biological signatures. We were introduced to some NASA researchers who had a wealth of interesting data and were seeking experts in ML and AI for scientific applications. After meeting with them, we started working on the data and developed models that we never imagined would be covered in the journal "Nature" as a story. We hope this develops into a long-term collaboration to explore even more topics.
What makes AI/ML and these agentic systems good tools for this type of research?
AA: Understanding the origins of life is such a complex problem that we thought AI would be a good tool. There are tasks, such as image detection, that are well-suited for the human brain, but this research is not like that. Understanding massive spectrometry data and integrating it with various modalities of data that exist in astrobiology is very challenging for humans. AI agents have shown promise in solving complex tasks, making it a natural fit for this problem.
DS: AI is also capable of going through large amounts of data much faster than a human would be able to. So, if datasets become very large, as they do with this project, it becomes much harder for human experts to parse through them.
With these large datasets, there’s also bound to be some junk or irrelevant data that the AI is effectively able to identify and disregard when doing the analysis.
(text and background only visible when logged in)
(text and background only visible when logged in)
Assistant Professor Amirali Aghazadeh (top) and Ph.D. candidate Daniel Saeedi.
A workflow chart showing how the AstroAgents system works. From data analyst to hypothesis and critical review, the agentic AI system is designed to mimic the scientific process.
(text and background only visible when logged in)
What were some of the challenges of this work?
DS: Going beyond the current literature was certainly a challenge. Also validating the hypotheses generated by the system was a big problem we had to overcome.
The origins of life are more abstract than other things, such as creating new drugs that involve agentic AI. You can test those. How do you test hypotheses for the origins of life? We had to rely heavily on our human experts to validate their findings, especially since we were seeking novel ideas. The validation of AI systems was a central topic at the International Conference on Learning Representations (ICLR) this year, where we presented AstroAgents.
Scientists have been attempting to understand the origins of life for a very long time. What did your models find as new potential hypotheses?
AA: Our goal was not just to generate any hypothesis. You can always generate simple or basic ideas, but we were looking to come up with some novel ideas where less research has been done. We were almost looking to surprise our collaborators at the NASA Goddard Space Flight Center.
AstroAgents generated more than 100 hypotheses, which our experts then rated on several criteria, including novelty, plausibility, and compatibility with the existing literature. We were pleasantly surprised at some of the scores our hypotheses got.
For example, our Gemini AI model analyzed chemical data from two better-known meteorites, Orgueil and Big Lew, and identified specific aromatic compounds. Based on their presence, the system proposed that these molecules likely formed in a distinct chemical environment within the early solar system, suggesting that the meteorites may have originated from a unique region with the right conditions for such compounds to develop.
(text and background only visible when logged in)
Our goal was not just to generate any hypothesis. You can always generate simple or basic ideas, but we were looking to come up with some novel ideas where less research has been done. We were almost looking to surprise our collaborators.
Amirali Aghazadeh
What is it like to have contributed not just ideas but substantial hypotheses with reasonable plausibility for future study?
AA: It was very exciting for me. This was an area of research I never imagined myself getting into. The last two years of this partnership with our collaborators have been very interesting, just learning about how much we don’t know about life and things beyond the Earth. It’s also been amazing to see AI bring something new to the field as well.
DS: It’s been one of the most interesting projects I’ve been involved with. It’s combined my AI agentic research with some of these big questions I’ve been asking since I was a child. It’s been an amazing experience, and I’ve learned a lot so far.
So, you were able to generate over 100 hypotheses, some good, some not so good. What’s the next phase of this project?
DS: We want to come up with a mechanism to validate the hypotheses. We can come up with lots of ideas, but we need them to have some kind of scientific validity to be taken seriously for further research to be done on them.
AA: Really, we are just scratching the surface on this project. We want to expand it in pretty much every way. Look at more data, analyze more literature, get the opinions of more experts, and generate more hypotheses. This paper we authored was a great starting point, but now we want to scale up to come up with more comprehensive results.
These models also have lots of applications outside of astrobiology. Some researchers have created drugs using agentic AI, so seeing where the technology goes will also be interesting.
(text and background only visible when logged in)
Related Content
NASA Honors Steffes with Exceptional Scientific Achievement Medal
The honor recognizes his pivotal role in unveiling the secrets of Jupiter's atmosphere through the Juno mission.
Space: The Current Frontier
ECE faculty are among the many the engineers whose technology is helping to work toward a new era of exploration and scientific discovery.