AI in space

How does AI find a planet nobody can see?

Telescopes now collect more data in one night than astronomers could look through in a lifetime. AI is the tool that sifts it, spotting tiny dips in starlight, faint moving dots and strange flickers that point to new worlds.

A planet passes in front of a star and the star's brightness graph dips PLANET CROSSES THE STAR →the star looks a tiny bit dimmer BRIGHTNESS OVER TIME (A “LIGHT CURVE”)the dip AI looks for
When a planet passes in front of its star, the star dims slightly. AI scans the brightness of hundreds of thousands of stars for that telltale dip.

For teachers

Grades 7–1245 minutesComputers or tabletsFree

Outcome: students classify real light curves and compare their own judgment with how AI does the same job.

What AI is doing right now

Four ways AI is exploring space

All four are real projects. Each one uses AI, often alongside physics-based software, to sift more data than people could ever check by hand.

01

Finding hidden planets

NASA’s Kepler telescope recorded the brightness of about 150,000 stars for years. AI learned what a planet’s dip looks like, then searched the data for ones humans missed.

Real exampleIn 2017, a Google AI found Kepler-90i, giving the star Kepler-90 eight known planets and tying our solar system’s record. In 2021, NASA’s ExoMiner validated 301 more planets at once: it judged that 301 of Kepler’s “planet candidates” were very likely real (a statistical check, not a new observation). NASA · NASA JPL

02

Spotting asteroids in old photos

Asteroids show up as faint dots that move a little between pictures. Algorithms can test thousands of possible paths to connect the dots across many images.

Real exampleIn 2024, the Asteroid Institute, using Google Cloud, ran a physics-based algorithm called THOR that linked about 27,500 candidate asteroids hiding in 400,000 old telescope images; an AI image classifier then helped decide which ones people should check first. Each one still had to be confirmed by the Minor Planet Center. Space.com

03

Sorting millions of alerts a night

The new Vera C. Rubin Observatory photographs the whole southern sky every few nights. Every change it sees becomes an alert, far too many for people to read.

Real exampleIn June 2026, Rubin officially began its 10-year survey. It sends out up to 7 million alerts a night, and automated “brokers” that use machine learning sort them so scientists can chase the interesting ones. University of Washington · Rubin Observatory

04

A student’s 1.5 million new leads

AI isn’t only for big labs. A high school student trained a model on old infrared telescope data to find objects that flicker, pulse or dim over time.

Real exampleMatteo Paz’s model flagged 1.5 million possible new variable objects in NEOWISE data. His paper appeared in The Astronomical Journal in 2024, and the project won him first place and $250,000 in the 2025 Regeneron Science Talent Search. University of Arizona

Try it yourself

Can you spot the planet?

Change the planet’s size and the amount of noise in the telescope data. Watch how hard the dip becomes to see, and why AI helps.

Transit simulator: a planet crosses its star and the star’s brightness graph dips

The line graph shows the star’s brightness over time. The dip is exaggerated so you can see it.

0.75%how much the star dims
Easyfor your eyes to spot
150kstars Kepler watched

Easy here, but notice the numbers: an Earth-sized planet dims a Sun-like star by less than 0.01%. That’s why astronomers rely on AI.

The pattern behind every AI tool

Data → pattern → prediction

Every “AI in…” page follows the same five steps. Learn them once and you can understand AI in any field.

1 · Data

Brightness of every star

Planet-hunting telescopes repeatedly measure the brightness of hundreds of thousands of stars. Kepler’s main data sampled about every 30 minutes for years; other missions and modes differ.

2 · Pattern

A small, repeating dip

A planet makes a short, U-shaped dip that repeats on a regular schedule.

3 · AI

Trained on known examples

AI learns from real planets and from look-alikes, such as two stars eclipsing each other.

4 · Check

People check

Astronomers check with other telescopes, or with careful statistics, before calling it a planet.

5 · Use

Choosing what to study

Confirmed planets become targets for telescopes like JWST to study their air.

Think like a scientist

What AI can’t do here

Knowing the limits is part of understanding the tool.

It can be fooled by look-alikes

Two stars orbiting each other, star spots or a camera glitch can make dips that resemble a planet.

It only knows what it was taught

AI trained on one telescope’s data may struggle with another’s. Truly new kinds of objects can slip past it.

It can’t confirm a discovery on its own

AI says “this looks like a planet.” Scientists then check with other telescopes, or with careful statistics, before a planet goes on the official list.

Most planets are invisible to this method

A planet only dims its star if its orbit lines up with our view. Most don’t.

Where people stay in charge: astronomers decide what to train the AI on, check its best candidates with other telescopes, and decide which discoveries are real.

Careers

Jobs where people work with this AI

Astronomy today is as much about code and data as telescopes. These jobs exist right now.

Data scientist at an observatory

Builds the software that turns raw telescope images into catalogs of stars, planets and asteroids.

Working with AITrains models to classify millions of objects and flag the strange ones.

Astronomer

Studies planets, stars and galaxies, and decides which questions are worth asking.

Working with AIUses AI to search huge surveys, then designs follow-up observations.

Planetary defense analyst

Tracks asteroids and works out whether any could hit Earth.

Working with AIUses orbit-linking algorithms to find faint moving objects early.

Telescope & instrument engineer

Designs cameras, detectors and spacecraft hardware.

Working with AIUses AI to predict failures and clean noise out of sensor data.

Classroom activity

Planet hunt with real NASA data

45 minutesGrades 7–12Computers or tabletsFree

Students become citizen scientists, searching real light curves from NASA’s TESS space telescope, then compare their eyes with how AI does the same job.

Teacher tip: try Planet Hunters TESS yourself the day before, to make sure it has light curves ready to classify.

  1. Warm up (5 min): Use the simulator above. Ask: what does a planet do to a star’s brightness?
  2. Hunt (20 min): In pairs, go to Planet Hunters TESS on Zooniverse and classify at least 15 light curves. Agree on each one before you submit, and mark any dips you see. Keep a tally: curves classified, curves with a dip, and curves you argued about. Some curves are simulated planets, and Zooniverse tells you when you have just classified one: note whether you had marked its dip. No account is needed; log in only if you want credit for a discovery.
  3. Tally (10 min): Add up the class totals on the board. Did any simulated planets come up, and how many of them did the class catch? Which kinds of curves caused the most arguments?
  4. Compare (10 min): Discuss: what made some dips easy and others hard? What would an AI need to learn to do this job?

Discussion questions

  • Why might people and AI each miss different planets?
  • If an AI finds 1,000 “candidates,” how should scientists decide which to check first?
  • Should an AI get credit for a discovery? Who should?

Fresh from the lab

Latest research in space

New studies are added every week and explained in plain language. Not every study uses AI, so ask yourself where AI could help. See all research →

New stories in this area are on the way. Browse all research →

Put curiosity to work

Read it. Try it. Question it.

Explore new AI research, or take a paper-first investigation into your classroom.