Case Study: Google DeepMind explores AI comedy writing with Prolific

Google DeepMind studies comedy with Prolific, scoring 54.6/100 for AI joke writing

A research team from Google DeepMind wanted to investigate how well large language models (LLMs) could serve as tools for comedy writing but faced the challenge of recruiting professional comedians experienced with AI. They also needed to create an environment for authentic evaluation and gather detailed feedback on complex issues like bias and censorship. To address this, they turned to the vendor Prolific for assistance.

Prolific provided a solution by successfully recruiting specialized comedy professionals and seamlessly organizing a series of workshops. This enabled the research team to gather rich qualitative and quantitative data. The results, gathered using Prolific's platform, revealed that comedians found AI-generated content bland and lacking essential human qualities for humor, with the LLMs receiving a mediocre score of 54.6 out of 100 on the Creativity Support Index, highlighting significant room for improvement.

Google DeepMind