GigSense: An LLM-Infused Tool for Workers Collective Intelligence

TL;DR

GigSense tackles the fragmentation and time constraints surrounding gig workers' collective problem-solving by integrating large language models with sensemaking theory in a modular, zoomable interface. The system supports rapid problem identification, diverse solution generation, and collaborative planning, while prioritizing human input and responsible AI use. In a controlled user study, GigSense yielded faster task completion, more identified problems, more or equally feasible solutions, and higher usability than a Dynamo-like control interface, with qualitative data highlighting enhanced sensemaking and worker solidarity. The work demonstrates the potential of LLM-infused, human-centered interfaces to expand inclusive collective intelligence in evolving work environments and outlines directions for responsible deployment and open-source development.

Abstract

Collective intelligence among gig workers yields considerable advantages, including improved information exchange, deeper social bonds, and stronger advocacy for better labor conditions. Especially as it enables workers to collaboratively pinpoint shared challenges and devise optimal strategies for addressing these issues. However, enabling collective intelligence remains challenging, as existing tools often overestimate gig workers' available time and uniformity in analytical reasoning. To overcome this, we introduce GigSense, a tool that leverages large language models alongside theories of collective intelligence and sensemaking. GigSense enables gig workers to rapidly understand and address shared challenges effectively, irrespective of their diverse backgrounds. Our user study showed that GigSense users outperformed those using a control interface in problem identification and generated solutions more quickly and of higher quality, with better usability experiences reported. GigSense not only empowers gig workers but also opens up new possibilities for supporting workers more broadly, demonstrating the potential of large language model interfaces to enhance collective intelligence efforts in the evolving workplace.

Figures (4)

• Figure 1: Overview of GigSense's problem-solving interface with the: A) "Data Gathering Module"; B) "Collaborative Solution Module" where workers can collaborate to create solutions to specific problems; C) "Problem Summary Module" where workers obtain summaries of specific problems to help their sensemaking; and D) "AI-Enhanced Solution Module" where workers obtain solution suggestions from LLMs.
• Figure 2: Gigsense's zoom-out and zoom-in view. These views can be accessed by gig workers using the corresponding buttons in the main interface. Gig-sense Zoomed-out view presents an overall view of the problems. Gig-sense Zoom-in view delivers a concise summary of the problem being analyzed, while also enabling users to explore relevant posts associated with that specific challenge.
• Figure 3: Overview of the control interface.
• Figure 4: Overview of: a) Median amount of time gig workers in each condition took to complete the different tasks; b)Box plot showing the evaluation of solutions proposed by gig workers in both groups based on their feasibility (on a 7-point Likert scale); c) Box plot showing task completion time in both groups; d) Box plot showing System Usability Scores in both groups