EECS doctoral student Senjuti Dutta is the first author of a paper that has been accepted in the Association for Computing Machinery’s CHI 2022. CHI is the premier international conference of human-computer interaction, and each year it brings together scholars and researchers from all over the world, as well as from various cultures, experiences, and positions.
“The conference’s overriding objective is to make the world a better place through interactive digital technology,” Dutta said.
Dutta’s research centers on human-computer interaction (HCI), crowdsourcing, and machine learning (ML). Her new paper, entitled “Mobilizing Crowdwork: A Systematic Assessment of the Mobile Usability of HITs,” focuses on a taxonomy of characteristics that define the mobile usability of human intelligence tasks (HITs), or tasks that require human intelligence to complete, for smartphones. The paper presents the iterative development of this taxonomy, highlighting the observed practices and preferences around mobile crowdwork.
“Crowdwork is an emerging style of work where anyone can engage with tasks that can be just about anything,” said Alex Williams, assistant professor and co-author. “Examples might include transcribing an audio file into machine-readable text, determining if an object is an image with a simple yes/no question, or even being asked to take a photo of a tree outside your apartment window. Today, crowdwork is facilitated by online platforms like Amazon Mechanical Turk, Crowdworkers, and Prolific that allow anyone in the public to complete these tasks in exchange for compensation.”
“The problem with modern crowdwork is that we have a mix of different task types that virtually everyone completes on the desktop,” Dutta said. “My work takes a step toward understanding a way that we can start moving work opportunities in crowdwork away from the desktop and toward the more mobile smartphone.”
They conducted a three-part study centered around this taxonomy and have identified a series of principles that enable researchers and practitioners in crowdwork to better design HITs for mobile experiences. The first phase of this study focused on designing the taxonomy through a series of deliberations guided by prior literature. The second phase validated the taxonomy’s correctness through a survey with workers on Amazon Mechanical Turk. The third phase demonstrated how the taxonomy can be used as a proxy for measuring HIT mobility by analyzing real HITs that were posted on Amazon Mechanical Turk.
Williams added that this new design taxonomy helps researchers and practitioners in crowdwork define, measure, and evaluate mobility.
“In the third phase of our study, we find that more than half of the 519 HITs that we sampled were not even accessible from a mobile device,” said Williams. “We applied our taxonomy to the remaining HITs that were accessible and were able to conclude that certain types of HITs are, by design, more conducive to mobile settings than others.”
Additional co-authors included EECS students Doug Lowe and Richard Rosenbalm, postdoctoral researcher Rhema P. Linder, and Assistant Professor Anastasia Kuzminykh of the University of Toronto, with whom Williams and Dutta collaborate closely.