Thesis by Yiftach Nagar: “This dissertation consists of three essays that advance our understanding of collective-intelligence: how it works, how it can be used, and how it can be augmented. I combine theoretical and empirical work, spanning qualitative inquiry, lab experiments, and design, exploring how novel ways of organizing, enabled by advancements in information technology, can help us work better, innovate, and solve complex problems.
The first essay offers a collective sensemaking model to explain structurational processes in online communities. I draw upon Weick’s model of sensemaking as committed-interpretation, which I ground in a qualitative inquiry into Wikipedia’s policy discussion pages, in attempt to explain how structuration emerges as interpretations are negotiated, and then committed through conversation. I argue that the wiki environment provides conditions that help commitments form, strengthen and diffuse, and that this, in turn, helps explain trends of stabilization observed in previous research.
In the second essay, we characterize a class of semi-structured prediction problems, where patterns are difficult to discern, data are difficult to quantify, and changes occur unexpectedly. Making correct predictions under these conditions can be extremely difficult, and is often associated with high stakes. We argue that in these settings, combining predictions from humans and models can outperform predictions made by groups of people, or computers. In laboratory experiments, we combined human and machine predictions, and find the combined predictions more accurate and more robust than predictions made by groups of only people or only machines.
The third essay addresses a critical bottleneck in open-innovation systems: reviewing and selecting the best submissions, in settings where submissions are complex intellectual artifacts whose evaluation require expertise. To aid expert reviewers, we offer a computational approach we developed and tested using data from the Climate CoLab – a large citizen science platform. Our models approximate expert decisions about the submissions with high accuracy, and their use can save review labor, and accelerate the review process….(More)”