Abstract: Volunteer-based crowdsourcing depend critically on maintaining the engagement of participants. We explore a methodology for extending engagement in citizen science by combining machine learning with intervention design. We first present a platform for using real-time predictions about forthcoming disengagement to guide interventions. Then we discuss a set of experiments with delivering different messages to users based on the proximity to the predicted time of disengagement. The messages address motivational factors that were found in prior studies to influence users’ engagements. We evaluate this approach on Galaxy Zoo, one of the largest citizen science application on the web, where we traced the behavior and contributions of thousands of users who received intervention messages over a period of a few months. We found sensitivity of the amount of user contributions to both the timing and nature of the message. Specifically, we found that a message emphasizing the helpfulness of individual users significantly increased users’ contributions when delivered according to predicted times of disengagement, but not when delivered at random times. The influence of the message on users’ contributions was more pronounced as additional user data was collected and made available to the classifier.

Citation: Avi Segal, Ya’akov Gal, Ece Kamar, Eric Horvitz, Alex Bower, Grant Miller. Intervention Strategies for Increasing Engagement in Volunteer-Based Crowdsourcing. International Joint Conference on Artificial Intelligence (IJCAI), New York, USA, July 2016.

Download: http://bit.ly/2jh9fgG

This work was presented at HAIDM 2015. The 2015 workshop on Human-Agent Interaction Design and Models was co-organised by SmartSociety.

Abstract: Crowdsourcing has emerged as a powerful way to provide computer systems with quick and easy access to human intelligence. However, there is a risk that online crowd workers could be directed to perform harmful tasks. To understand the impact of financial incentives on paid crowd workers’ willingness to behave maliciously, we conducted a series of experiments in which we hired crowd workers via one crowdsourcing task (Attack task) to attack a different crowdsourcing task (Target task. We found that roughly one third of all crowd workers were willing to provide the attack task with potentially sensitive information from the target task, and that we could double this number by increasing the payment of the Attack task. Based on exit interviews and community feedback, we discuss some of what workers reported. Our findings reveal a measurable cost to completing malicious work that well-meaning task designers can leverage to protect their systems from attack.

Citation: Walter Lasecki, Jaime Teevan and Ece Kamar. What’s Your Price? The Cost of Asking Crowd Workers to Behave Maliciously.

Download: http://bit.ly/1VB1DTh

This work was presented at HAIDM 2015. The 2015 workshop on Human-Agent Interaction Design and Models was co-organised by SmartSociety.

Abstract: It can be hard to automatically identify sensitive content in images or other media because significant context is often necessary to interpret noisy content and complex notions of sensitivity. Online crowds can help computers interpret information that cannot be understood algorithmically. However, systems that use this approach can unwittingly show workers information that should remain private. For instance, images sent to the crowd may accidentally include faces or geographic identifiers in the background, and information pertaining to a task (e.g., the amount of a bill) may appear alongside private information (e.g., an account number). This paper introduces an approach for using crowds to filter information from sensory data that should remain private, while retaining information needed to complete a specified task. The pyramid workflow that we introduce allows crowd workers to identify private information while never having complete access to the (potentially private) information they are filtering. Our approach is flexible, easily configurable, and can protect user information in settings where automated approaches fail. Our experiments with 4685 crowd workers show that it performs significantly better than previous approaches.

Citation: Walter Lasecki, Mitchell Gordon, Jaime Teevan, Ece Kamar and Jeffrey Bigham. CrowdMask: Privacy-Preserving Crowd-Powered Systems.

Download: http://bit.ly/1YF37Ke