The light of every star is its unique passport, but the entries are written in the language of spectra and colors. For old giants, such documents are almost indistinguishable: their spectra overlap, and deciphering the details by hand was practically impossible. The Gaia space telescope, equipped with high-precision photometry — the measurement of stars' brightness and color, collected millions of these light fingerprints. Then a neural network, trained on a vast dataset, learned to read age and mass from them, much as an experienced translator deciphers an ancient text from barely visible strokes — for instance, from subtle absorption lines in the spectrum.

The big surprise: the program, without any hints, independently figured out how to separate reddening by cosmic dust — microscopic solid particles in the interstellar medium from the actual cooling of the star. This is the physical law of interstellar extinction, which was not originally built into it. The neural network uncovered it by analyzing statistical patterns in the data: that is, it 'understood' that part of the redshift is caused not by the star's temperature but by light scattering off dust grains. Moreover, it restores erased lines in damaged spectra — graphs of energy distribution across wavelengths and can even generate new 'passports' for fictional stars, predicting what their light would look like.

Satellite photometry, thus, turns into stellar archaeology: with 'passports' for billions of stars, astronomers can reconstruct the history of the Milky Way, our galaxy, how it grew by absorbing smaller galaxies and survived collisions. The path to this discovery began in the 19th century, when Joseph von Fraunhofer first spotted dark lines in sunlight, those very spectral fingerprints of elements, and a century later Cecilia Payne-Gaposchkin proved that stars are mostly made of hydrogen, overturning ideas about the composition of the cosmos. Their legacy now lives not only in textbooks but also in silicon chips and machine learning algorithms.