Friday, March 11th

Explaining Things Differently: A Crowdsourcing


ApproachJoseph Williams

Guest Speaker: Jacob Whitehill and Joseph Jay Williams
MIT Campus

This is a two-part talk about crowdsourcing novel explanations from ordinary people and other students themselves, as well as automated mechanisms to decide which explanation to offer to a new student, so as to help learners learn as effectively as possible.

In the first part, Jacob Whitehill will present joint work with Margo Seltzer (Harvard SEAS) on crowdsourcing 400+ video-based explanations from ordinary people on Amazon Mechanical Turk on introductory concepts relating to logarithms — e.g., “How do you solve the equation log 5x = 2?” Preliminary findings suggest that (a) short 5-10 minute explanatory videos can help learners significantly improve their understanding of basic logarithm concepts; and (b) a large variety of explanations — in terms of approach, format, and quality — can be efficiently collected by crowdsourcing from ordinary people across the world.

In the second part, Joseph Jay Williams will present a system for crowdsourcing written explanations from students while they are solving problems in EdX and Canvas. These explanations are provided to future students, who rate how helpful the explanations are for learning. A machine learning algorithm (Thompson Sampling) uses Bayesian statistics to automatically increase or decrease how often explanations are presented, based on how highly they are rated, and to add new explanations as students produce them. This approach to simultaneously crowdsourcing explanations and helping people learn is shown to have a positive impact on learning, and produce explanations that satisfy the needs of instructional designers.

Recording available at