How might AI accidentally save the world?

Imagine a situation: you are the scientist who wants to create a robot using unsupervised learning. At first, you want him to be smart, so you spend two weeks putting in all the data. Then, you ask him to write articles about global warming in physical copy. It understands everything now so what it needs to do is to write and write and write. You keep it working and go to eat your dinner.