Imagine driving a car, but every time you turn the wheel, the car reacts ten minutes later.
You are no longer driving. You are praying.
That is exactly the problem with space probes that travel far from Earth. And it is why an American start-up, AstroForge, has decided to hand over the controls to an artificial intelligence.
The problem: distance
AstroForge wants to go and mine asteroids, those big rocks floating in space that sometimes contain rare metals.
But an asteroid is far away. So far that a radio signal — which travels at the speed of light — takes several minutes to make the round trip.
If the probe detects a problem, sends the information to Earth, an engineer thinks about it and sends back an order, a good while has passed. In the meantime, the probe has kept moving. For a delicate manoeuvre near an asteroid, that is far too slow.
The probe therefore has to be able to manage on its own.
The solution: "Solo"
AstroForge has built its own AI, named Solo. It learned from data from around 2,500 spacecraft sensors: temperature, power, position, orientation.
Its role: to monitor all of that continuously, understand what is happening, and decide what to do, without waiting for anyone.
It does not completely replace the conventional programs that have been piloting probes for decades. It sits on top, like a conductor coordinating all the instruments.
Before letting Solo pilot for real, AstroForge will fly it in "ghost mode". At the end of 2026, it will be on board a first probe. It will make every decision... but will not apply them. Engineers will compare what it wanted to do with what was actually done. It is exactly like driving with a learner: the pupil holds the wheel, but the instructor has a brake pedal. If Solo passes this test, it will pilot a second probe on its own in 2027.
Why this is a real turning point
We are used to hearing about AIs that answer questions or write texts. If they get it wrong, we reread, correct, start again.
This is different. In space, there is no one to reread. No undo button. No tow truck. If Solo makes a bad decision, the probe is lost, along with the tens of millions it cost.
This is probably one of the first times an AI of this kind will be truly alone. With no human able to intervene in time.
What could go wrong
An AI learns from examples. Solo has seen thousands of situations recorded by sensors. But space holds surprises that were in no recording.
Faced with something genuinely new, an AI can react unpredictably, and with a great deal of confidence. That is the main risk, and it is exactly why the "ghost mode" phase is so important: it makes it possible to spot those moments before they cost a probe.
What we take away
This project says something simple: as soon as you go far enough, humans can no longer keep up in real time. It is not a question of intelligence, it is a question of physics. Light does not go any faster.
The further we explore, the more we will have to trust machines that decide on their own. And the real question is not whether they are capable of it. It is how we verify that they are before sending them where no one will be able to catch them.