Rethinking Research Crowdsourcing:
Human Feedback Meets LLM-Powered Digital Twins
Built With:
React + GPT-4o + Azure
In response to challenges presented by covert AI usage in the crowd working process, we propose and evaluate a hybrid framework using digital twins - personalized AI models that collaborate with workers while preserving human agency when the AI is uncertain.
Equal contributions were made to this research by Amanda Chan, Cathy Di, Joseph Rupertus, and Gary Smith. It also would not be possible without the guidance of Varun Rao, Manoel Horta Ribeiro, and Andrés Monroy-Hernandez.
System Highlights:
System Design that Celebrates the Harmony of Humans and AI
– Recruited 88 crowd workers to test the system, along with 9 researcher and crowd worker interviews to gather feedback on the design.
– Strong reactions revealed exciting possibilities for scalable, ethical, and authentic crowd work.
Participant Platform that Empowers and Delights Crowd Workers
– Developed from the ground up with a React frontend, Azure Static Web App backend, and OpenAI's GPT-4o chat model.
– Seamlessly integrates AI, detecting uncertainty and redirecting necessary questions to humans to preserve agency.
– Carefully crafted to support crowd workers in completing surveys with ease, mindfulness, and real-time progress tracking.
Curious about something, or just want to get in touch?
Connect with Me
