Answering my own question: Looks like research has been done on this.
From the paper "Skype me! Socially Contingent Interactions Help Toddlers Learn Language" by Sarah Roseberry, Kathy Hirsh-Pasek, and Roberta Michnick Golinkoff:
"Our findings suggest that language learning occurs in the socially contingent interactions made possible via video chat."
Here socially contingent means not just timing speaking to the babies reactions, but customizing the content to their context (using the names of the child's pets and siblings for example). Per the authors exact description:
Troseth and colleagues (2006) defined contingent interactions as a two-way exchange in which the adult on video established herself as relevant and interactive by referring to the child by name and by asking children specific questions about their siblings and pets. This view of social contingency posits that socially contingent interactions should be appropriate in content (Bornstein, Tamis-LeMonda, Hahn & Haynes, 2008) and intensity (Gergely & Watson, 1996). It is a departure from a narrower definition of contingency, which focuses solely on timing and reliability (Beebe et al., 2011; Catmur, 2011).
The authors also seem to agree with my friend's comment that just sitting your kid in front of the TV doesn't do much. Researchers have termed this the "video deficit."
The authors also note that toddlers specifically appear to benefit more from watching two people converse on screen compared to Blue's Clue's style talking directly to the child type video.
Although our findings support the utility of video chats, they also hint at how adept children are at distinguishing real contingency from other types of interactions. Our results from the yoked video condition indicate that simply posing questions to children and pausing for the answer did not result in language learning if the children were not able to interact contingently with the person on video. In fact, it may be more beneficial for young children to witness two characters interacting with each other on screen than for the characters to attempt to talk with children directly. Toddlers seem to learn better from watching a social interaction on video than from being directly addressed through video (O'Doherty, Troseth, Shimpi, Goldenberg, Akhtar & Saylor, 2011).