Revealing Abstract Arts with Feedback Induced Crowdsourcing to LLM Sourcing

Abstract

Crowdsourcing has received a major attention in solving creative tasks in recent times. Some creative problems are so abstract that they require repetitive interaction between the requester and crowd worker. Moreover, defining a ground truth for such creative tasks is a challenge. This paper aims to address the problem of revealing the content of abstract arts with a feedback-induced mechanism -- both via crowdsourcing and LLM sourcing. As abstract arts are interpreted in different ways, it is interesting to elucidate the content through interaction. We propose an approach that employs a corrective feedback mechanism to enable the requester and crowd workers to interact. The effectiveness of this approach is demonstrated by annotating 30 abstract arts on a crowdsourcing platform. The results show that feedback motivates workers to provide detailed responses, interact further with the requester, and reveal more on the abstract content. Further sentiment analysis on the discussion data reflects the importance of corrective feedback in crowdsourcing. We further extend this by outsourcing the tasks to LLMs and observed a better output. However, some interesting challenges like hallucination and ethical participation by the LLMs emerge through this.