AI Hunting Other Planets

How Artificial Intelligence Is Processing the Vast Data of the Planet Search

In the summer of 2016, a team of astronomers at Google and NASA's Ames Research Center trained a neural network on a set of light curves from the Kepler space telescope. Within months, it had identified two new exoplanets that human reviewers had missed — worlds that had been hiding in data already analysed, validated, and set aside.

The discovery made headlines not because the planets were unusual, but because of what found them. A machine had looked at the sky and seen something we had not.

That moment was not an anomaly. It was a symptom. The field of exoplanet astronomy — the search for planets orbiting stars beyond our own Sun — had entered a new era, one defined less by the power of telescopes than by the power of the algorithms trained to read their output.

We had built instruments capable of detecting the faintest whisper of a planet's gravitational pull, or the hairline dimming of a star as a world crossed its face. What we had not built, until recently, was the analytical capacity to hear everything those instruments were telling us.

Until AI