Imagine a high school senior in California doing what professional astronomers chase for their entire careers: unveiling over 1.5 million celestial bodies that were hiding in plain sight, all thanks to an AI tool he created himself. This isn't just a cool story—it's reshaping how we peek into the infrared cosmos, and it's sparking debates on who gets to explore the stars. But here's where it gets controversial: could a teen's fresh perspective outshine the so-called experts?
Matteo Paz didn't luck into this breakthrough. With deliberate skill, he crafted a custom machine learning system to comb through mountains of untouched NASA data from the NEOWISE mission. This involved sifting through more than 200 billion data points from over a decade of infrared sky scans, hunting for faint clues like flickers, pulses, and dimmings that point to cosmic wonders. Think of these as subtle changes in brightness over time, revealing things like quasars—those ultra-bright galactic cores powered by supermassive black holes—or supernovae, the explosive deaths of massive stars. There are also eclipsing binaries, where two stars orbit each other and periodically block each other's light, and other variable phenomena that are tricky to spot but crucial for understanding the universe. These signals were missed before, partly because NEOWISE, launched in 2009 to hunt near-Earth asteroids, amassed a huge archive of data but left much of it unexplored for dynamic events that evolve over time.
NEOWISE excelled at gathering info, much like other space missions that collect vast datasets but defer the heavy lifting of analysis to later generations. Enter Paz, a 17-year-old with a love for stars and coding skills, who joined Caltech's Planet Finder Academy—a program mentoring young talents in advanced astrophysics. His guide, Davy Kirkpatrick, a top scientist at Caltech's Infrared Processing and Analysis Center (IPAC), initially thought they'd manually check a tiny slice of the data. But Paz pitched a full algorithmic approach to handle it all.
The outcome? An AI model fusing Fourier transforms—which break down signals into frequency components to spot patterns—with wavelet analysis, a method that zooms in on details at different scales. This lets it catch brightness shifts in distant objects, often signaling unseen cosmic activities. Kirkpatrick was floored: 'Matteo's model started delivering fascinating findings right away, and it only improved from there.'
Scanning the heavens, one subtle pulse at a time, Paz's versatile algorithm detects light variations spanning hours to years, uncovering millions of light curves—graphs of brightness over time—that eluded experts and standard tools. Some changes are quick flashes, others slow fades that slipped through the cracks. And this isn't just eye candy for stargazers; these discoveries could sharpen our grasp of stellar life cycles, chart galaxy layouts, or pinpoint rare happenings like gravitational lensing, where a massive object bends light from a distant source, creating a natural telescope. Some might even uncover wholly new types of phenomena, ripe for scrutiny by upcoming observatories such as the Vera Rubin Observatory or NASA's James Webb Space Telescope.
The complete list of 1.5 million objects is slated for public release in 2025, while scientists are already buzzing over early findings. Paz's paper in The Astronomical Journal is fueling talks on the hidden power of old data, especially when mixed with new algorithms and outside viewpoints. And this is the part most people miss: what if legacy space data holds answers to today's biggest questions, but only if we're willing to think differently?
Paz's innovation shines even brighter for its broad potential. Built to spot stellar variability, his pipeline can adapt to any time-based data series. 'Once you see it's about tracking change over time, the doors swing wide open,' Paz explained in a Caltech chat. For instance, it could trace pollution trends in the environment, decode earthquake patterns in seismic records, or even follow stock market shifts. Time-series analysis on reliable datasets like NEOWISE is booming across fields, and NASA is ramping up AI in astronomy, collaborating with IPAC experts. Paz's project proves machine learning isn't just a speed boost—it's a game-changer when trained well.
At the heart of this tale is mentorship and opportunity. Kirkpatrick, an expert in brown dwarfs and faint stars, pushes to include diverse voices in astronomy. 'We've got tons of data but too few eyes on it,' he notes. 'With talent like Matteo's, it's vital to provide the backing they need.' Paz was ready: enrolled in Pasadena Unified's Math Academy, a public program fast-tracking gifted math students through college courses, he nailed AP Calculus BC by eighth grade and, by senior year, authored a solo paper in a top journal.
This story raises big questions: Should we rethink how we allocate resources in science, trusting young innovators over established pros? Or is this just a lucky anomaly? But here's where it gets controversial—does democratizing data access mean anyone with a computer could redefine our universe, or should we stick to the 'experts'? What do you think—could AI like Paz's level the playing field, or does it risk overlooking human intuition? Share your thoughts in the comments; I'd love to hear if you agree, disagree, or have your own take on how technology is changing discovery!
