Testing is hard for many reasons. A big reason is that AI is non-deterministic; that means, it is so complex and has so many interactions that it is no practical to try to explicitly model every possible state. However, computer chips and operating systems are non-deterministic, so probabilistic methods exist, they just often aren’t applied to robotics. Instead, customers depend on demonstrations, such as in the short story Robinc. Demonstrations are often just for validation, to show that system does what the customer intended or wanted. But it is important to verify that the system really, truly works correctly. Another reason why testing is hard is because there are three ways the robot can fail: some external force (like a mine explosion), a failure of the robot hardware or software, or a human-robot interaction failure. Typically manufacturers concentrate on preventing hardware failures because software failures are harder, and more time consuming, to detect. Learning adds a twist- both because there’s the problem of whether the learning algorithm is programmed correctly and whether it will learn the right things! Testing is not a sexy topic so it doesn’t show up much in fiction. Learn more about testing in Robotics Through Science Fiction: Artificial Intelligence Explained Through Six Classic Robot Short Stories.