A new Workflow for Human-AI Collaboration in Citizen Science.

The unprecedented growth of online citizen science projects provides growing opportunities for the public to participate in scientific discoveries. Nevertheless, volunteers typically make only a few contributions before exiting the system. Thus a significant challenge to such systems is increasing the capacity and efficiency of volunteers without hindering their motivation and engagement. To address this challenge, we study the role of incorporating collaborative agents in the existing workflow of a citizen science project for the purpose of increasing the capacity and efficiency of these systems, while maintaining the motivation of participants in the system. Our new enhanced workflow combines human-machine collaboration in two ways: Humans can aid the machine in solving more difficult tasks with high information value, while the machine can facilitate human engagement by generating motivational messages that emphasize different aspects of human-machine collaboration. We implemented this workflow in a study comprising thousands of volunteers in Galaxy Zoo, one of the largest citizen science projects on the web. Volunteers could choose to use the enhanced workflow or the existing workflow in which users did not receive motivational messages, and tasks were allocated to volunteers sequentially without regard to information value. We found that the volunteers working in the enhanced workflow were more productive than those volunteers who worked in the existing workflow, without incurring a loss in the quality of their contributions. Additionally, in the enhanced workflow, the type of messages used had a profound effect on volunteer performance. Our work demonstrates the importance of varying human-machine collaboration models in citizen science.