In 2020, machine learning will become a more integral part of our lives, able to perform ever more tasks. Unfortunately, we still cannot trust it to make important decisions. The fact is that machine learning can be fooled. For example, by wearing glasses that are coated in a special way you can cheat a facial recognition algorithm and impersonate someone else. But by 2050 we will have no choice but to trust machine learning everywhere, for example, in self-driving vehicles. In many areas where machine learning outperforms people, it turns out that people are smarter and can easily cheat the system. In general, there are fewer victims, but now a "hacker" can kill. In other areas of machine learning applications, the situation is similar.