Unnatural pedagogy : a computational analysis of children's learning to learn from other people.

Abstract

DedicationTo Baxter Eaves Sr.One could not ask for a more loving, supportive, well-mustachioed dad.iii Acknowledgments First and foremost, I would like to express my sincerest appreciation to Patrick Shafto, whose support and guidance over the years has made all of this work possible. I am indebted to Dr. Shafto not only for the effort he has put into developing this work, but for the effort he has put into developing me as a scientist and as a professional.Dan Navarro and Amy Perfors were major collaborators on the work that forms the foundation of this work and Thomas Griffiths and Naomi Feldman were instrumental to the work on infant-directed speech. I would also like to thank my lab comrades: Nick Searcy, Kelley Durkin, Asheley Landrum, and Campbell Rightmyer who have provided valuable feedback on the many projects I have conducted over the years that comprise this work. such powerful knowledge accumulators. These innate assumptions result in developmental patterns observed in epistemic trust research. This research seeks to create a computational account of the development of these abilities. We argue that pedagogy is not innate, but rather that people learn to learn from others. We employ novel computational models to show that there is sufficient data early on from which infants may learn that people choose data pedagogically, that the development of children's epistemic trust is primarily a result of their decreasing beliefs that all informants are helpful, and that innate pedagogy would not lead to more rapid learning. We connect results from the pedagogy and epistemic trust literatures across tasks and development, showing that these are different manifestations of the same underlying abilities, and show that pedagogy need not be innate to have powerful implications for learning.v