Mountain View's SETI Institute lets AI agents steer its alien-signal telescope

The Mountain View-based SETI Institute and the Breakthrough Listen project are using AI agents to help run the Allen Telescope Array, their radio telescope in Northern California, chip maker NVIDIA said in an Oct. 30 blog post.

What the Allen Telescope Array is

The array stands at Hat Creek Radio Observatory, about 300 miles north of San Francisco near Lassen Peak.  It has 42 dish antennas, each 6 meters (20 feet) across, and was the first radio telescope designed from the ground up to search for signals from other civilizations, the SETI Institute says.

A radio telescope collects radio waves instead of visible light.  The array listens across about 1 to 14 gigahertz.  Combining the signals from all its dishes gives it "as many pixels on the sky as a dish that is 1000 feet across," the institute says, and its receivers could detect a cellphone on the moon.  Besides hunting for alien signals, it studies pulsars, flashes called fast radio bursts, exploding stars and flares on other stars.

UC Berkeley and the institute dedicated the array on Oct. 11, 2007, with the first 42 of a planned 350 dishes.  It had cost $50 million by then, half from the foundation of Microsoft co-founder Paul Allen, Berkeley said at the time.

What the AI agents do

A researcher can type one request in plain language, and software agents steer the antennas, calibrate the instrument, catch and fix data errors, run AI models on incoming radio signals and plot the results.  NVIDIA, based in Santa Clara, says the system handles 2.4 terabits of data per second on its chips, processing it as it arrives instead of storing it first.

But this isn't something to fear.  Rather, it's a breakthrough in translation.

Will AI make the telescope more productive?

The value comes in two parts.  The first is faster analysis of the flood of incoming data, and that part has been measured.  In a peer-reviewed 2025 study in Astronomy & Astrophysics, a SETI Institute AI system at the array spotted fast radio bursts 600 times faster than the older software, which needed 59 seconds to process 16.3 seconds of data.  It was also 7% more accurate and raised about 10 times fewer false alarms, the institute said.  Faster than real time means events can be caught as they happen.

The second is the feedback loop:  agents that read the data and then change the telescope's and software's settings themselves.  NVIDIA says agents "compress what would take humans weeks or years into hours," but neither company has published an independent measure of how much more science that yields.

NVIDIA says the setup has already made a real-time AI detection of a pulsar, a spinning dead star, in the Crab Nebula.  No alien signal has been reported.

Source:  NVIDIA blog, "Agentic AI Joins the Search for Extraterrestrial Signals," Oct. 30, 2026; SETI Institute, Allen Telescope Array page and Nov. 5, 2025 release; Astronomy & Astrophysics, October 2025; UC Berkeley release, Oct. 11, 2007.