Using AI to Give People a Voice: A Case Study in Michigan — AI • Objectives • Institute
This text reports on a case study of an AI-based community outreach project by the AI Objectives Institute. Key takeaways include:.
Project Description:.
- “Talk to the City” (TttC), an AI analytics platform.
- Analyzing video interviews with former inmates on the challenges of reintegration
- Worked with Silent Cry, a non-profit organization
Main Results:.
- AI extracts key assertions, topics, and stories from interviews
- Identify common issues such as digital literacy, employment, and housing discrimination
- Facilitates direct communication to policy makers
AI Applications and Challenges:.
- Identify topics and subtopics using GPT-4 Turbo
- Accurate categorization of claims, avoidance of duplication, and maintaining context for personal stories are challenges
Improvements:.
- Distinguishing between interesting and obvious content
- Remove duplicate claims
- Explanation of High Context Reference
- Error detection and correction
Future Prospects:.
- Ease of use of the tool and improved clarity of reporting
- WhatsApp integration, geographic and demographic metadata integration
- Iterative Discussion Tracking
Conclusion:.
- Demonstrated potential to leverage AI to amplify the voices of under-resourced communities
- Able to create reports that bring the human story to the forefront
- Room for improvement, but could be a powerful platform for advocacy
This project demonstrates the potential of AI to increase understanding of important social issues and amplify the voices of underrepresented groups.
The project’s information gathering methods are as follows
Participants:.
- For 12 participants
- Ten of them are former inmates (Returning Citizens).
- All live in Michigan.
Interview Methods:.
- Video interviews using Google Meet
- July 8 to August 7, 2023.
- Video collection for a total of approximately 8 hours
Interviews:.
- Questions about challenges facing the community
- Hear ideas for solutions (e.g., policy measures) to the problem
Data processing
- Transcribe video using Descript
- Remove filler words (“ah”, “eh”, etc.)
- Interviewer’s statement also deleted.
- Upload edited video to Vimeo
Consent process:.
- Obtain consent from each participant to publish their name and interview online
- Also obtain consent to input anonymized transcriptions into the LLM (Large Language Model)
- Describes LLM in non-technical terms
- Clarify that transcripts are uploaded to OpenAI’s servers
Review Process:.
- Each participant reviews the generated report
- Flags inaccurate, misclassified, or improper claims
- Internal review of flagged claims, edit or delete
AI Applications:.
- Create accurate time-stamped transcriptions using Whisper (speech recognition AI)
- Identify topics and subtopics using GPT-4 Turbo
- Extract key assertions from each interview using LLM
- Classify each claim into appropriate topics and subtopics using LLM
This approach combines the power of AI with human review and approval processes to ensure accurate and ethical data collection and analysis.
Heal Michigan is a report generated using video transcripts from interviews with activists in Michigan, and providing links to the source videos. Interviews conducted in partnership with Silent Cry (coming soon).
relevance 663ce2cdaff09e00004428a6
This page is auto-translated from [/nishio/Heal Michigan](https://scrapbox.io/nishio/Heal Michigan) using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.