Google AI Space Telescope Projects has been nominated for the 2026 World AI Awards in the Aerospace category, recognising work that has demonstrated how machine learning can help scientists extract discoveries from enormous volumes of astronomical data.
Google does not operate the Kepler Space Telescope. That distinction matters. Kepler was a NASA mission designed to search the Milky Way for planets beyond our solar system. Google researchers instead developed machine-learning techniques capable of examining data already collected by the telescope and identifying planetary signals that conventional searches had missed.
The World AI Awards recognises organisations, individuals, products and technologies contributing to the development and real-world application of artificial intelligence across industries.
For Google AI, astronomy presented a problem well suited to machine learning: telescopes can produce far more observations than researchers can reasonably inspect by hand. Finding something important means separating tiny signals from an enormous amount of noise.
Teaching AI to hunt for planets
NASA’s Kepler telescope searched for exoplanets by watching stars for tiny, periodic drops in brightness. A planet passing between its star and the telescope can produce that dip, known as a transit.
The difficulty is that planets are not the only things that produce suspicious signals.
Binary stars, starspots, cosmic rays and instrumental noise can create patterns that resemble planetary transits. Kepler generated tens of thousands of possible signals requiring classification, while weaker signals could fall below traditional detection thresholds.
Google researcher Christopher Shallue and astrophysicist Andrew Vanderburg approached the problem with deep learning.
They trained a convolutional neural network using around 15,000 previously classified Kepler signals, including approximately 3,500 verified planets or strong planet candidates. Instead of programming fixed rules for what a planet should look like, the model learned to distinguish likely planetary transits from false positives using examples.
Then they pointed it toward data that deserved another look.
A neural network finds Kepler-90i
The experiment produced more than a promising research paper.
Google’s neural network helped identify Kepler-90i, a rocky planet orbiting Kepler-90, a Sun-like star approximately 2,545 light-years from Earth. The discovery gave the Kepler-90 system eight known planets, tying our solar system at the time for the largest known number of planets orbiting a single star.
The researchers also identified Kepler-80g, another previously overlooked planet.
Their published model achieved a 98.8% ranking result on its test set when distinguishing plausible planetary signals from false positives. That figure describes performance within the researchers’ evaluation rather than a blanket accuracy rate for discovering planets across all astronomical data.
Google later released the processing and neural-network code publicly, giving other researchers a starting point for applying similar techniques to astronomical datasets.
The real problem is no longer collecting data
Modern astronomy has an unusual challenge. Better instruments do not simply reveal more of the universe; they produce an extraordinary amount of information that somebody — or something — has to examine.
Kepler observed roughly 200,000 stars over four years, generating about 14 billion data points, according to Google. Searching all the possible orbital combinations manually would be impractical.
That is where AI becomes interesting for astronomy.
Machine learning does not replace the telescope. It changes what researchers can recover from the telescope’s observations.
Google has continued exploring that idea beyond Kepler. Research published in 2025 demonstrated how Gemini could classify astronomical transient events such as exploding stars using examples from the Pan-STARRS, MeerLICHT and ATLAS surveys. Google reported 93% accuracy across three datasets after providing the model only 15 annotated examples for each survey.
It is a different problem from exoplanet hunting, but the underlying pressure is familiar: astronomical instruments are generating signals faster than humans can comfortably sort through them.
Graham Cooke, President of the World AI Awards, said:
“Google AI’s work with space telescope data demonstrates one of the most compelling opportunities for artificial intelligence in scientific discovery: helping researchers find meaningful signals inside datasets that have become too large to examine manually.
“The discoveries associated with NASA’s Kepler data show that AI can give scientists another way to revisit observations and uncover patterns that might otherwise remain hidden. We congratulate Google AI Space Telescope Projects on its 2026 World AI Awards nomination and look forward to seeing how these techniques contribute to the next generation of astronomical research.”
Google AI Space Telescope Projects joins organisations, researchers, entrepreneurs and technology developers being recognised through the 2026 World AI Awards.
The programme recognises organisations, individuals and technologies contributing to the development and application of artificial intelligence across industries.
Astronomy offers a particularly revealing case. The telescope remains the instrument collecting light from distant stars. Scientists still determine whether the evidence supports a discovery. AI sits somewhere in between, helping researchers decide where, among billions of measurements, they should look next.
The Kepler work showed that this approach could produce something tangible: two planets hiding inside data humanity had already collected.
And that may be the more interesting story. Sometimes a better view of the universe does not require another telescope. It requires finding more in the data already sitting on Earth.
Learn more about Google AI and its research at ai.google.
Discover the World AI Awards 2026, explore the nominees and learn more about the awards at worldawards.ai.
