Integrating citizen science with online learning to ask better questions

Department of Pediatrics, UC San Diego {vipandey, srk}@eng.ucsd.edu

{amamir, jdebelius, ehyde, tkosciolek, robknight}@ucsd.edu

Abstract Online learners spend millions of hours per year testing their new skills on assignments with known answers. This paper explores whether framing research questions as as- signments with unknown answers helps learners generate novel, useful, and difficult-to-find knowledge while increas- ing their motivation by contributing to a larger goal. Collab- orating with the American Gut Project, the world’s largest crowdfunded citizen science project, we deploy Gut Instinct to allow novices to generate hypotheses about the constitu- tion of the human gut microbiome. The tool enables online learners to explore learning material about the microbiome and create their own theories around causal variances for microbiome. Building on crowdsourcing or serious games that use people as replaceable units, this work-in-progress lays our plans for how people (a) use their personal knowledge (b) towards solving a larger real-world goal (c) that can provide potential benefits to them. We hope to demonstrate that Gut Instinct citizen scientists generate use- ful hypotheses, perform better on learning tasks than tradi- tional MOOC learners, and are better engaged with the learning material.
Can online learners perform useful work in citizen science projects? Crowdsourcing scales well and provides good results when people’s untrained intuitions are on average good, e.g. in tasks labeling images (von Ahn et al. 2004) and performing real-time captioning (Lasecki et al. 2012). This holds for tasks most people are naturally expert at, such as recogniz- ing objects in images, or transcribing what’s spoken in their language (Surowiecki 2005). However, for many tasks, people might have lousy estimates or guesses, if any. Such tasks require domain-specific expertise in breadth of knowledge (such as identifying a cat’s breed in an image) or in understanding deeper features (such as describing the quality of a painting). In such cases, crowdsourcing tries to do useful work by training novices but the results are mixed.

Citizen Science projects, though important, appeal to a limited set of hobbyists Citizen science seeks to solve large scientific challenges using a distributed set of people to perform tasks (Bonney et al. 2009). Biology problems dominate popular online citizen science efforts, such as Foldit (https://fold.it ) for protein folding, EteRNA (www.eternagame.org/ ) for RNA design, and Phylo (phylo.cs.mcgill.ca/) for small-scale multiple sequence alignment problems. Moreover, scien- tific datasets created from massive efforts like the Human Genome/Microbiome Projects are difficult to analyze due to (a) vast set of features and (b) gaps in our understanding of these topics. This interest in finding alternate ways to analyze data works well with people’s native expertise in tasks such as identifying high-level patterns, used in games like Phylo. Designing learning modules for citizen science has demonstrated improved domain knowledge among participants (Lee et al. 2016).

However, most citizen science projects still provide mini- mal training and utilize participation towards low- cognition tasks like identifying certain objects in images. Since these topics from niche area, they interest hobbyists and do not scale to people beyond a small community. Galaxy Zoo (www.galaxyzoo.org ) is such an example where space enthusiasts help classify galaxies. Recent citi- zen science projects, such as American Gut Project (http://americangut.org/ ) have pulled people in the loop as contributors: subjects who provide their own physical and behavioral data. We consider the next step of this evolu- tion. How can we transform excited contributors into ac- tive collaborators who can generate hypotheses as well? Our key insight is that motivated contributors to a citizen science project can develop expertise using online learning material and collaboratively create novel knowledge.

Online learning is underexplored as a platform to bring together crowds to do useful work Online learners spend considerable time learning new skills and testing them on assignments with known an- swers. Could we better support their learning by asking them to apply their skills and fresh perspective towards citizen science problems with unknown answers? We test our idea in the context of the human gut microbiome re- search. The human gut microbiome is the community of microbes (and their gene products) interacting in the hu- man gut. The American Gut Project (AGP) gives people the ability to contribute to microbiome science by