Agentic AI and the Search for Extraterrestrial Life

Minutes

On May 29, 2026, Members of the Society and guests joined the speaker for a reception and dinner at 5:45 PM in the Members’ Dining Room at the Cosmos Club. Thereafter they joined other attendees in the Powell Auditorium for the lecture proceedings. In the Powell Auditorium of the Cosmos Club in Washington, D.C., President Larry Millstein called the lecture portion of the 2,536th meeting of the Society and the 95th Joseph Henry lecture to order at 8:08 p.m. ET. He began by welcoming attendees, thanking sponsors for their support, announcing new members, and inviting guests to join the society. Scott Mathews then read the minutes of the previous meeting which included the lecture by Harris Lewin, titled “The Earth BioGenome Project: Reading the Library of Life”. The minutes were approved as read.

President Millstein then introduced the speakers for the evening, Michael Garrett, of the University of Manchester, and Adam Thompson, of NVIDIA. Their lecture was titled “Agentic AI and the Search for Extraterrestrial Life”.

The first speaker was Michael Garrett, whose portion of the lecture focused on the astrophysical foundations of the Search for Extraterrestrial Intelligence (or SETI). Garret began by saying that SETI was looking for “technosignatures”; anomalies in astronomical data that might be the telltale signs of intelligence in the Universe. He outlined how SETI has transitioned from a philosophical pursuit into a rigorous, data-driven observational science, highlighting how the scale of next-generation radio telescopes create a critical need for advanced AI. He said that the massive datasets produced by these instruments will require new technologies and new techniques in order to detect technosignatures. He said that this transformation was largely driven by the explosive growth in exoplanet discoveries over the last two decades, confirming that planets are ubiquitous in our galaxy, and many reside in the "habitable zone" of their host stars.

Garrett gave specific examples of the types of technosignatures SETI is searching for, now and in the near future. These included: Narrow-band radio transmissions (the traditional focus of SETI), Optical or Laser Pulses (brief, intense flashes of light used for interstellar communication or propulsion), Megastructures (Anomalous transits or thermal signatures that could indicate massive engineering projects like Dyson spheres), and Broad-band radio “leakage” (aggregate radio emissions resulting from a large number of devices, operating over a wide range of frequencies). He introduced the acronym BRaT, for broadband radio technosignatures. He emphasized the need for AI in identifying BRaT’s, distinguishing them from the background of billions of confounding, faint, radio sources in the Universe.

To find these elusive signatures, Garrett detailed the astronomical community's roadmap for scaling up its observational power. He categorized the hardware evolution into three phases: current upgrades (e.g. the Allen Telescope Array, the Very Large Array, and the MeerKAT Radio Telescope), next-generation radio telescopes (e.g. the Square Kilometer Array, and the Next-Generation Very Large Array), and future concepts, like Lunar Farside Observatories.

Garrett ended his portion of the lecture by quoting Frank Drake, saying “At this very minute, with almost absolute certainty, radio waves sent forth by other intelligent civilizations are falling on the Earth.”

The second portion of the lecture was presented by Adam Thompson, who focused on the computational architecture required to solve the exabyte-scale data bottleneck in modern radio astronomy. He explained how NVIDIA and the SETI Institute are utilizing Edge Supercomputing and Agentic AI, directly at the sensor source, to bypass bandwidth limitations, perform real-time signal processing, and autonomously isolate technosignatures from background noise.

Thompson said that the digitization of analog RF signals creates data streams at rates of hundreds of gigabits or even terabits per second. He said that it is economically and physically impossible to pipe this raw data over standard networks to centralized servers for analysis. Rather than moving the data to the compute, the compute must be moved to the data. This requires deploying high-performance computing clusters directly at the observatory, or “on the edge."

Thompson outlined the strategy of applying Edge Supercomputing: streaming data directly into GPU’s, bypassing the CPU’s, processing datasets immediately upon generation, discarding the noise and forwarding only the relevant anomalies, reducing the required bandwidth by orders of magnitude. He discussed the hardware associated with this type of edge-computing, including: DSP’s (digital signal processors), GPU’s (graphics processing units), and FPGA’s (field programmable gate arrays). He described some of the algorithms or operations performed by such hardware, including: fast Fourier transforms, polyphase filtering, real-time RF interference mitigation, and anomaly detection. Thompson highlighted some of the advantages of shifting these workloads to GPU’s. These included: highly flexible, software defined processing, rapid deployment of algorithms, dynamic scaling, massive bandwidths, and the integration of AI models and signal processing on the same devices.

Thompson said that because SETI is searching for the unknown, traditional machine learning is insufficient, because it requires labeled training data. Thompson explained that Agentic AI in this application allows for unsupervised anomaly detection, dynamic noise suppression, and autonomous telescope operation. If an anomalous signal is detected, the AI can immediately command the telescope to dwell on the source and track it in real time.

Thompson ended his portion of the lecture by saying that “…NVIDIA is a platform company. We make tools that other people can build on top of, and then use for their own science.”

The lecture was followed by a Question and Answer session.

A member asked about the impact of AI on the training and education of young people in this field. Garrett responded that he sees many students using AI on a regular basis, and that he was “…a little worried about that.” He said that while he thinks AI is a fantastic tool, dramatically increasing his productivity, he wonders whether younger people will build-up the experience and familiarity with data on a deep level.

A member on the live stream asked if a more advanced civilization, sending radio signals to Earth, could compensate for or change the Doppler shift of the transmitted signal. Garrett responded that other civilizations could be thousands of years ahead of us in technology, and that as a result, the AI’s looking for technosignatures must be open to detecting anomalies that we would not expect.

A member asked about “model collapse”, the idea that AI generated data is fed back into an AI as training data, ultimately causing the model to “generate gibberish”. Thompson responded that NVIDIA seeks to “connect world class models to world class instruments”. He said that he does not have the experience to discuss what happens when AI goes wrong. Garrett responded that astronomy deals with dynamic datasets: the sky is constantly changing. He said that this ensures that AI models are constantly changing and evolving, and are therefore less susceptible to “degeneration”.

A guest asked about the possibility of transferring specific algorithms or processes from the GPU’s to other hardware, like FPGA’s. Thompson responded “…the most realistic and probably best systems are heterogeneous compute”, meaning that some combination of GPU’s with CPU’s and FPGA’s will likely be required. He gave the example of an agentic AI in high energy particle physics that requires nanosecond latency, saying that GPU’s are currently not capable of such speeds, but FPGA’s are.

After the question and answer period, President Millstein thanked the speakers and presented them with PSW rosettes, signed copies of the announcement of their talk, and signed copies of Volume 17 of the PSW Bulletin. He then announced speakers of up-coming lectures and made a number of housekeeping announcements. He adjourned the 2,536th meeting of the society at 10:22 pm ET.

Temperature in Washington, DC: 22.8° Celsius
Weather: Fair

Dinner attendance: 74
Lecture attendance:
In person: 126
Live Stream: 40
For a total of 166 viewers
Views of the video in the first two weeks: 330

Respectfully submitted, Scott Mathews: Recording Secretary