Search for: All records

This paper studies the design of an AI tool that supports gig knowledge workers, rather than displacing them, focusing on text-based generative AI technologies. Through a formative study involving interviews and design activities, gig workers shared their views on text-based generative AI and envisioned applications where AI acts as managers, secretaries, and communication aids. Leveraging these insights, we created a generative-AI enhanced tool, Office-Mind AI, to aid gig workers. Our research advances the conversation around algorithmic labor by designing a worker-focused intelligent tool. This tool harness collective intelligence among workers and AI, fostering productive human-AI partnerships. We conclude by discussing the future prospects of collective intelligence tools designed for worker-AI collaborations. 

Citizen science games must balance task difficulty with player skill to ensure optimal engagement and performance. This issue has been previously addressed via player-level matchmaking, a dynamic difficulty adjustment method in which player and level ratings are used to present levels best suited for players' individual abilities. However, this work has been done in small, isolated test games and left out potential techniques that could further improve player performance. Therefore, we examined the effects of player-level matchmaking in Foldit, a live citizen science game. An experiment with 221 players demonstrated that dynamic matchmaking approaches led to significantly more levels completed, as well as a more challenging highest level completed, compared to random level ordering, but not greater than a static approach. We conclude that player-level matchmaking is worth consideration in the context of live citizen science games, potentially paired with other dynamic difficulty adjustment methods. 

Citizen science projects that rely on human computation can attempt to solicit volunteers or use paid microwork platforms such as Amazon Mechanical Turk. To better understand these approaches, this paper analyzes crowdsourced image label data sourced from an environmental justice project looking at wetland loss off the coast of Louisiana. This retrospective analysis identifies key differences between the two populations: while Mechanical Turk workers are accessible, cost-efficient, and rate more images than volunteers (on average), their labels are of lower quality, whereas volunteers can achieve high accuracy with comparably few votes. Volunteer organizations can also interface with the educational or outreach goals of an organization in ways that the limited context of microwork prevents.