It began with a problem of scale.
Modern telescopes like the Vera C. Rubin Observatory, set to begin operations in October 2025, are poised to revolutionize astronomy by capturing unprecedented volumes of data—millions of time-variable astrophysical events across the sky each night. But with this scale comes a new challenge: How do you process, classify, and interpret this firehose of information when human analysis alone can’t keep up with the hundreds of new events being reported by Rubin every second?
That’s the frontier being navigated by Gautham Narayan, professor of astronomy at the University of Illinois and deputy director for astrophysics research at the NSF-Simons AI Institute for the Sky (SkAI). His group has spent over a decade pioneering artificial intelligence in astrophysics, developing early classification and inference algorithms, and helping lay the groundwork for a supercharged field by SkAI. “My group has been doing AI and astrophysics for about the last decade,” Narayan explains. “SkAI is a way to take the things I've been doing locally here at Illinois, and really make it a much larger effort that is multi-institute.”
Founded to harness the parallel revolutions in AI and astronomy, the SkAI Institute is working toward what Narayan calls “foundation models for astronomy.” These are AI models inspired by systems like ChatGPT but tailored to the complexities of “multi-modal” astrophysical data—images, spectra, and time series. If applied at scale, these models could be transformative. “If you could do classification, forecasting, inference, and analysis now at an industrial scale for billions of objects in a year,” he says, “that has the potential to change everything we know about astrophysics.”
A powerful example of this potential came with GW170817—the first observed merger of two neutron stars. Astronomers detected the event across multiple wavelengths and combined data from observatories worldwide to form a multidimensional view of the explosion and the origins of heavy elements. “What was valuable was seeing how combining data from all of these telescopes could give you such a multidimensional view,” Narayan recalls. “The obvious question is, well, what if you could do that for 100,000 objects?”
That’s where AI steps in—not just as a tool for managing data at scale but as an engine for discovery.
A mission at the intersection
SkAI’s mission is to marry astronomy and AI in a scientifically rigorous and fundamentally interpretable way. “We care about our model’s predictive power,” Narayan emphasizes, “and making sure our models are interpretable, that a human can look at what an AI algorithm has done and say, ‘Yes, this makes physical sense.’”
The Institute’s reach is broad and collaborative, with partner institutions including Northwestern, the University of Illinois, the University of Chicago, the Adler Planetarium, Fermilab, Argonne National Lab, and community colleges like Parkland in Champaign. Cultural institutions such as the Spurlock and Krannert Art museums on campus are also involved, broadening the impact beyond traditional scientific spaces.
Education and outreach are key pillars of SkAI’s vision. Students at Illinois whose work aligns with the Institute’s priorities can receive funding, and the same opportunities extend to postdoctoral fellows, including through a unique preceptorship program. “We also have opportunities for postdocs,” Narayan notes, “including a preceptor postdoctoral fellow who has to be engaged in education and outreach work in the community colleges.”
Rethinking astronomy education and infrastructure
Looking ahead, Narayan envisions a research environment where scientists interact with powerful AI tools not by writing specialized code but through intuitive, natural-language queries. “You could ask it: ‘Can you find me 100 low-redshift spiral galaxies in Rubin images with space-based Euclid images that also show stellar streams?’” he says. “Something that would have taken a grad student a year, you could potentially do in seconds.”
This shift could significantly reshape how educators teach astronomy. “I think in the long term, we are going to have to revise our astronomy curriculum to embed AI into almost every class we offer,” Narayan predicts.
The ripple effects extend well beyond university classrooms. Through efforts like SCiMMA (Scalable Cyberinfrastructure for Multi-Messenger Astrophysics), Narayan and his collaborators are bridging siloed observatories nationwide, building a national infrastructure where telescopes and AI systems work in concert. “SCiMMA’s mission is to connect all of these things,” he explains, “so that ultimately you can build this AI-enabled future that I’ve been trying to describe.”
What’s next?
Narayan is building a foundation model for time series data, especially for explosive events like supernovae and kilonovae. The goal is to move beyond simply classifying these events to analyzing them at scale—studying their physical properties, distributions, host environments, and behaviors as a population. “That’s a hard task because nobody’s done it systematically yet,” he says. “It’s hard to assemble the data and develop methods that combine data from different facilities with traditional non-AI techniques. But I want to get to the point where we can essentially ask these AI models for a large sample and analyze them all at once.”
In parallel, he continues work on projects like the Rubin Observatory, the Young Supernova Experiment, and SCiMMA—all feeding into SkAI’s broader mission of building an AI-powered ecosystem for astronomy. “I think you’ll see a revolutionary change in how we deal with data, not just doing what we do now at a larger scale, but asking entirely new questions that we couldn’t have asked before,” he adds.