Artificial intelligence has taken a bold step into astrobiology with the development of a new agentic AI system, “AstroAgents,” designed to autonomously conduct scientific research into the origins of life in the Universe. As reported by Nature, the system has already produced over 100 scientific hypotheses and is set to assist NASA in analyzing samples from Mars.
Developed by a team of researchers including astrobiologist Denise Buckner from NASA’s Goddard Space Flight Center and computer scientist Amirali Aghazadeh from the Georgia Institute of Technology, AstroAgents comprises eight specialized AI agents. Each is tasked with functions such as data analysis, planning, hypothesis generation, and critique — forming a collaborative “scientist team” driven by large language models (LLMs).
“It’s helping us build a better understanding of how molecules form in space, how molecules form from life on Earth and how they’re preserved — and then which specific signs should we be searching for,” said Buckner.
The team introduced AstroAgents on April 27 at the International Conference on Learning Representations in Singapore. According to Nature, the AI system was tested on mass spectrometry data from eight meteorites and ten Earth soil samples, including those from Antarctica and the Atacama Desert. Using two different LLMs — Claude Sonnet 3.5 and Gemini 2.0 Flash — the AI conducted ten iterative rounds of analysis, generating 101 hypotheses from Gemini and 48 from Claude.
While Claude’s results were clearer and less error-prone, Gemini’s hypotheses included 24 considered “novel” by Buckner, who individually evaluated each. One notable idea from Gemini suggested that certain Earth molecules could act as biomarkers indicating extraterrestrial life.
However, some experts remain cautious. Astrobiologist Michael Wong of Carnegie Science pointed out that only one person evaluated the hypotheses, noting that a broader peer review would add credibility. “Even the hypotheses that received high scores did not teach me anything new about the mystery of life’s emergence,” Wong said.
Still, the potential of agentic AI — systems that independently decide, adapt, and evaluate — is exciting researchers. “We realized that because of the complexity of the data, it’s better for the agent to assign multiple tasks to multiple ‘scientists’,” said Aghazadeh. He described the planner’s role in distributing tasks among the agents as “the magic of the system.”
Looking ahead, AstroAgents is expected to play a key role in analyzing samples from Mars, particularly from an ancient lakebed that NASA plans to explore. These samples, set to arrive on Earth in the 2030s, could hold crucial clues to whether life once existed on the red planet.
Despite lingering questions about originality and validation, Aghazadeh remains optimistic. “We are just starting and scratching the surface,” he said. “Agentic AI will play a key part in understanding the origin of life.”
