AI-Generated Materials for Dark Matter Detection
The universe is a mysterious place, and it’s largely composed of mysterious stuff. All of the stuff we can see—galaxies, stars, planets, trees, people—are made of ordinary matter, but this only accounts for about 5 percent of what’s out there. The rest is some combination of so-called “dark matter” and “dark energy.” Scientists have inferred the existence of these mysterious dark entities based on calculations of their gravitational impact on the parts of the universe we can see but have yet to directly observe them.
Carlos Blanco, assistant professor of physics and a co-hire of the Institute for Computational and Data Sciences at Penn State, is using AI to develop materials with properties that will hopefully increase our ability to detect dark matter.
“Dark matter doesn’t emit, reflect, or absorb light, and the only way that we know it interacts with ordinary matter is through its gravitational impact,” said Blanco. “So, we are trying to develop dark matter detectors that could pick up any faint signals that result from dark matter colliding with an ordinary atom. One way to do this is by identifying materials that might be more likely to produce these signals in ways that we can interpret.”
Current dark matter detectors contain some sort of sensor, usually buried deep underground. If a particle of dark matter collides with an atom in the sensor, it might produce a faint signal in the form of light or an excited electron. These events are rare and the signals weak, so the problem becomes how to distinguish a real signal indicating dark matter interacting with the detector and noise in the system resulting from other types of interactions.
“The next generation of dark matter detectors have to be cleverer about how they distinguish an actual dark matter signal from the massive amount of noise inherent in these searches,” said Blanco. “One way to do this is if we can determine the directionality of the interaction—did the dark matter particle come in from above the detector, from the side? So, the problem becomes one of materials science. Can we identify a material that would give us this information? Current detectors have been built with materials that are basically off-the-shelf parts, we want some that is purpose built.”
To identify these materials, Blanco uses machine learning and generative AI. He is building a model, similar to a large-language model, trained on a large database of small molecules—those with around 30 atoms or less—that have some known properties that indicate how sensitive they may be as detectors. The model “learns” how these properties are related to the structure and composition of the molecules in the training set and can suggest new molecules to try to optimize their ability to detect dark matter. It is analogous to how ChatGPT, for example, is trained on existing text, then can produce original sentences.
“As we try to understand the composition and evolution of the universe, our calculations tell us that we are missing as much as 80 percent of the matter; things like galaxies behave like they are five times more massive than what we can detect,” said Blanco. “One of the central problems in particle physics and cosmology is trying to figure out the nature of these missing ingredients in the universe. I studied chemistry as an undergrad and shifted to physics during grad school, but it turns out that this combination of interests allowed me to carve out a niche in this field. Part of the reason I came to Penn State, is that its strengths in materials science make it a playground for folks like me.”
Editor's Note: This story is part of a larger feature about artificial intelligence developed for the Winter 2026 issue of the Eberly College of Science Science Journal.
