Aligning Generalisation Between Humans and Machines