I can see the potential for descrimination based on a machine learned perceived identity based on web searches and online activity. Imagine not being able to go into a store or be allowed in neighborhoods because of research on certain kinds of diseases or comments made in forums in the past or views supposedly or actually held.
I consider this a pretty phenomenal article. At a time where most articles suggesting we should fear AI imagine unrealistic Skynets and hyperintelligent robotic species, this highlights problems that are almost certain to occur in the future.