Intelligent Tutoring Systems are often modeled after human tutors; however, the effectiveness of this strategy is yet to be determined. Research on media interactions suggests that behaviors with humans are similar to those with computers. Intelligent Tutoring System studies have said the opposite. In this study we compared a human-human and a human-computer tutoring system in terms of metacognitive, social, and nonsense statements to dig deeper into these interactions. We discovered that the interactions were quite different between human-human and human-computer tutoring. With a human, participants expressed more positive metacognitive statements and social statements. When interacting with a computer tutor, students were more likely to make negative metacognitive statements and social statements. In addition, the interpretation of these results differed between the two corpora. In human-human tutoring, the more often a participant made positive metacognitive statements, the worse their learning gain. Their social dialogue had no impact on learning gain. In human-computer tutoring, the more negative and positive metacognitive statements and the more negative social statements they gave the worse their learning gain. It is clear from this study that students do not act the same with a human tutor as they do with a computer tutor. Therefore, designers of ITS systems should not just blindly model their systems after human tutors. The differences in human and computer interactions should also be considered.