What's Actually Happening in a Tutoring Environment?

So, given that we're wanting to simulate human one-on-one tutoring, or get as close as we can to it to achieve the 2-sigma effect, what is it about the tutoring experience, exactly, that we want to replicate?

So far, the Chi article has been the most compelling for me.

Chi, Michelene et al (2001). Learning from Human Tutoring. Cognitive Science 25, 471-533 (Here's a link to the PDF)

This study used a control group with no tutoring, a group receiving normal tutoring, and a group that received "suppressed tutoring" in which the tutors did not provide information and answers but did still provide an interactive environment. The students who received suppressed tutoring still progressed just as well as those who received normal tutoring.

So it's not the tutor as a repository of knowledge that makes a difference. It appears, on the surface at least, that it doesn't even matter how much the tutor knows. A student can still excel when paired with a tutor who doesn't provide answers but does provide questions - provides an interactive platform on which students can construct their own knowledge. (Did I just say "construct their own knowledge?" Dangit. That's one of those cliché phrases I promised never to say.)

I found two other articles that report on an AI tool called AutoTutor which attempts to replicate actual human tutoring.

Graesser, A., Wiemer-Hastings, K., Wiemer-Hastings, P., Kreuz, R., & the Tutoring Research Group. (2000). AutoTutor: A simulation of a human tutor. Journal of Cognitive Systems Research, 1, 35-51.

Graesser, A., VanLehn, K., Rosé, CP., Jordan, PW., Harter, D. (2001). Intelligent tutoring systems with conversational dialogue. AI Magazine, 22(4), 39-52. (Here's a link to the PDF)

The articles don't report any sort of testing on the tool and, quite frankly, I remain rather dubious after reading them, but they do bring up one interesting point that goes along with this train of thought.

"We discussed three projects that have several similarities. AUTOTUTOR, ATLAS, and WHY2 all endorse the idea that students learn best if they construct knowledge themselves. Thus, their dialogues try to elicit knowledge from the student by asking leading questions. They only tell the student the knowledge as a last resort."

While I don't know that AutoTutor really does a very good job of creating an opportunity for students to construct knowledge themselves, there is still the assumption that what's really going on here doesn't have as much to do with the tutor as it does with the student. The reason students flourish in a one-on-one environment is because of the student half of the one-. The student needs to be questioning, interacting, involved. So does it matter who's on the other side of the table?

Is this just, then, a pedagogy question? I'm increasing tempted to think it is.