[...] Daphna ran through the same nine sequences with all the children, but with one group, she acted as if she were clueless about the toy. ("Wow, look at this toy. I wonder how it works? Let's try this," she said.) With the other group, she acted like a teacher. ("Here's how my toy works.") When she acted clueless, many of the children figured out the most intelligent way of getting the toy to play music (performing just the two key actions, something Daphna had not demonstrated). But when Daphna acted like a teacher, the children imitated her exactly, rather than discovering the more intelligent and more novel two-action solution.
These experts in machine learning argue that learning from teachers first requires you to learn about teachers. For example, if you know how teachers work, you tend to assume that they are trying to be informative. When the teacher in the tube-toy experiment doesn't go looking for hidden features inside the tubes, the learner unconsciously thinks: "She's a teacher. If there were something interesting in there, she would have showed it to me." These assumptions lead children to narrow in, and to consider just the specific information a teacher provides. Without a teacher present, children look for a much wider range of information and consider a greater range of options.
This experiment is from:
D. Buchsbaum, A. Gopnik, T.L. Griffiths, and P. Shafto (2011). Children's imitation of causal action sequences is influenced by statistical and pedagogical evidence. Cognition (in press). pdf
The other paper cited in the Slate article is:
E. Bonawitz, P. Shafto, H. Gweon, N.D. Goodman, E. Spelke, and L. Schulz (2011). The double-edged sword of pedagogy: Instruction limits spontaneous exploration and discovery. Cognition (in press). pdf