But algorithms, unlike humans, are susceptible to a specific type of problem called an “adversarial example.” These are specially designed optical illusions that fool computers into doing ...