jiminmun / particip-ai

Particip-AI

[Paper]

Overview

This data contains survey answers from 295 demographically diverse, US-based participants on AI usage. More specifically, the survey contained three following sections.

1. Use Cases: Participants were asked to brainstorm use cases of AI and choose a use case that might be most drastically changed.
2. Harms and Benefits: Participants to write harms and benefits of both developing and not developing the use case in free-text format, identify the groups impacted most, and rate the scale of impact.
3. Application Decision: Participants were asked to make a decision on whether the application should be developed.

Structure

In data folder

• codes: Folders of codes and their definition (code-def-0107v1.json), fewshot examples used for prompting (fewshot-0107v1.json), and prompt template for applying codes with an LLM (prompt-template-0107v1.json)
  • task-coding: Q7
  • benefits-coding: Q8_4
  • harms-coding: Q11_4, Q11_5, Q14_4, Q14_5
  • not-dev-benefits-coding: Q20_4
  • not-dev-harms-coding: Q17_4
  • groups-coding: Q9, Q12, Q15, Q18, Q21
• coding_results: Coding results from LLM code application
• human_annotations: Human applied coding results of 80 samples
• themes: High level themes of each code
• raw_data.csv: Answer to all questions, including demographics and AI literacy questions

Additionally, all questions are in questions.json. See survey.pdf for survey format presented to participants.

Shield:

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.