A small spacecraft lands on Jupiter's moon Europa and sits calmly on the frozen surface, figuring out what's normal. After a year has passed, a plume of water suddenly breaks through the surface ice, and the AI brain in the spacecraft will notice that it contains organic carbon.

We won't have to tell it what to do. It will make its own decisions. Its brain will say, "Something unusual is happening." It will move to the plume, observing and thinking, then make all the right choices, which experiments to run and what data to collect. If a living microbe has been forced out from the under-ice oceans, our AI rover will be the first to know, and we won't find out that life exists there until it chooses to send a message back to Earth. 

Artificially intelligent spacecraft may seem like the stuff of science fiction, some dream of Star Trek Voyager fans, but not to Steve Chien and Kiri Wagstaff of NASA's Jet Propulsion Laboratory. Writing in Science Robotics, they stress the need for conceiving and building spacecraft that don't wait for humans to tell them what to do.

In or about 2022, NASA will launch the Europa Clipper, an orbiting scientific laboratory that will focus on Europa. It will not stay in orbit above the moon, but will circle Jupiter and pass over Europa at least 45 times as close as 10 miles. It will also drop a lander onto the surface. But 2022 may be too soon to place really powerful AI on the mission. Even JUICE, the Jupiter Icy Moons Explorer, will have to be told what do to by ground control millions of miles away.

But it won't be long, say Chien and Wagstaff, before powerful AI will be ready to put into orbiters, landers, or comet probes. The AI will be able to discern discrepant events--a plume of water, a rock that is chemically odd, a geological formation that is out of place--and know to act.

Rudimentary AI programming is already in use. The Mars Opportunity WATCH program allowed the rover to spot new dust devils on the surface and examine them. The AEGIS program allows ChemCam to use its laser on rocks without human controller telling it to do so.

The orbiter Earth-Observing 1 can spot sulfur on glaciers, and the Autonomous Sciencecraft Experiment recognizes fires, floods, and volcanic eruptions without being directed to do so.

We may be at the beginning of merging AI with space exploration. The vision of Chien and Wagstaff, laid out in Robotic Space Exploration Agents, goes far beyond Europa, Ganymede, or Callisto (another possible site for life). They imagine power AI missions beyond the solar system, on comets, even to the stars.