The speaker explains how they look at simple features like words and length of material when interpreting data, but notes that they don't know why users are clicking on certain things or reading certain information without taking into account their prior models.