Abstract
In this case study, we describe the design and implementation of the Backyard Worlds: Cool Neighbors citizen science project, which combines image-level deep learning with Zooniverse-hosted online crowdsourcing to mine large astronomical sky maps for rare celestial objects called “brown dwarfs.” Specifically, Cool Neighbors uses machine learning to pre-select the sky images shown to volunteers. Cool Neighbors represents an excellent opportunity to interrogate the effects of incorporating artificial intelligence into a citizen science project; its sibling project, Backyard Worlds: Planet 9, uses no artificial intelligence, providing a natural point of comparison for participant engagement metrics. Through analysis of more than 10 million total Zooniverse classifications from the combination of Cool Neighbors and Backyard Worlds: Planet 9, among other results, we find (1) Cool Neighbors volunteers perform ~3x more classifications per unit of time invested than Backyard Worlds: Planet 9 volunteers, and (2) each registered Cool Neighbors participant performs ~2–5x more classifications than each registered Backyard Worlds: Planet 9 participant. We also discuss our measured approach to presenting the complementarity of machine learning and citizen science in volunteer-facing Cool Neighbors materials. Finally, we present a survey of advanced Backyard Worlds participants, which indicates that these citizen scientists are by and large not dissuaded from participating in Cool Neighbors because of its usage of artificial intelligence.