An orbital carrier controlled by artificial intelligence could be used to patrol and counter attacks in space , according to a new study by Chinese scientists. They say a large orbital platform carrying hundreds of cubesats – tiny satellites that weigh about 1kg (2.2lbs) – could defend China’s space assets with speed and efficiency. But they say it would need help from AI to determine exactly when and where to release the cubesats so they could fend off enemy satellites. According to the researchers, the complexity of a large and fast space battle would be beyond the human brain – and even beyond some powerful AI algorithms. Studying the best strategy for AI to control an orbital carrier would have “strong economic and military value”, the team said in a paper published in Chinese Space Science and Technology , a peer-reviewed journal run by the China Academy of Space Technology, on June 25. The research was led by Zhang Jin, a professor with the College of Aerospace Science and Engineering at the National University of Defence Technology in Changsha. China has alleged that SpaceX Starlink satellites came dangerously close to the new Chinese space station on two occasions last year, and raised concern that the scale of these unfriendly encounters could increase in the near future. Months later, Chinese and US satellites had a game of “geostationary orbit cat and mouse”, according to a Space News report on June 16, which said such encounters were becoming more frequent. Military researchers in May called for a plan to disable or destroy SpaceX’s Starlink satellites if they threatened China’s national security. Zhang and his team said an orbital platform carrying cubesats could be used to patrol and defend against any organised and continuous attack in space. Nasa’s Webb telescope shows where stars are born and how they die They proposed using AI for mission planning by using it to answer key questions such as the direction of orbit transfer, when the cubesats should be released, and the timing of encounters with other satellites. The researchers came up with a way to do this, built on a simulation model. Their “multi-round greedy search” method is an algorithm designed to command four orbital platforms to inspect nine hostile targets in less than a day. They put it to the test under a high-precision orbit model and also compared it with a hybrid encoding genetic algorithm – one of the most popular optimisation methods. Their algorithm was found to be 227 times faster than the genetic algorithm – in 20 rounds of testing, it found the best result in four minutes. The genetic algorithm found rough solutions in 200 minutes, and better results took 900 minutes. The scientists said this all came down to a key difference in strategies – theirs was more focused on the big picture while the genetic algorithm spent a lot of time and resources on the finer details. The greedy algorithm deals with multiple constraints but uses low-precision parameters at first, and when it finds an acceptable solution it skips the higher precision calculations. Zhang said this was found to be a more efficient approach than traditional optimisation methods. The AI could also give humans a choice of approaches to take. According to the paper, the algorithm was able to plot a mission that used the least fuel, offering a route that would cost 96kg (212lbs) of fuel and take 68 hours; it also suggested the shortest mission time that would cost 950kg of fuel and take 18 hours. How our growing belief in AI self-awareness is becoming a problem “In the future, we will add randomness to the search strategy to overcome the limitations of the greedy algorithm and obtain global optimal results,” Zhang said in the paper. They said an orbital carrier using AI could also be used for other purposes, such as in-orbit refuelling and maintenance. In April, another team of scientists in China said they had developed AI that could use tactics like deception to hunt satellites.