Submissions:2025/Wikipedia editing + AI: Could there be a right balance?
This submission has been noted and is pending review for WikiConference North America 2025.
Title:
- Wikipedia editing + AI: Could there be a right balance?
Type of session:
- Edit-a-thons or Workshops (30-120 min)
Session theme(s):
- Credibility, Future of Wikipedia
Abstract:
This meetup/workshop will focus on the challenges and opportunities of using generative AI (GAI) to assist Wikipedia editing. Our initial experiments and observations reveal a possible correspondence between editing experience and GAI tool effectiveness. Inexperienced editors may often accept problematic outputs containing hallucinations, omissions, AI slop, and AI verborrhea (especially when it comes to models in “research” mode). Experienced editors, however, tend to use GAI assistance to enhance their topic comprehension, reduce initial inertia when starting articles, adopt more engaged and analytical editing approaches, and achieve increased productive output. We are very interested in hearing about other editors' Wiki-GAI experiences and perhaps brainstorming process ideas together to better understand if/how different experience levels can effectively leverage these tools.
We plan to touch on a number of our own questions as well as those of the participants: How do we balance possible efficiency gains from GAI with Wikipedia's commitment to human knowledge curation? How might GAI assistance affect the recruitment and retention of new editors? What are the implications for Wikipedia's reliability if readers cannot distinguish between human and AI contributions? What are the practical implications of attempting to ban GAI use entirely, and given the difficulty of proving that content has been generated by AI, how would enforcement affect community dynamics and trust? What would effective transparency and regulation frameworks actually look like in practice? Should we devise training or certification to help editors use GAI tools effectively? Finally, would it be worthwhile to develop Wikipedia-specific open-source GAI that inherently follows community policies and procedures?*
Participants are encouraged to review two AI-generated articles on deaf culture in New York City before attending the session as a means to have tangible examples for our discussion. These example articles were generated using two prompts: first, the models were asked to generate a list of articles about deaf culture in New York City, then to use the generated list to create a Wikipedia article on the topic following Wikipedia's neutrality guidelines and sourcing requirements.
- Example article generated by Claude Sonnet 4 with Research: https://claude.ai/public/artifacts/1319551c-ff2a-4520-a7d2-4171c64a35dd
- Example article generated by Gemini 2.5 Pro with Deep Research: https://docs.google.com/document/d/1Om-xRlqMle7EaWKmGu2ZOAwQfHHO21YTHLR7eoqQ_rA/edit?usp=sharing
Ximena Gallardo C. (User: Doctorxgc) has practiced Critical AI methods since 2022 and she recently taught an undergraduate level class on investigating GAI for research writing. She has written two Wikipedia pieces with GAI assistance: the first draft of the article for the International Salsa Museum using NotebookLM and the history section of the article Artificial intelligence in politics using Claude Sonnet 4 Pro with Research.
Richard Knipel (User: Pharos) is the first Wikipedian to use GAI (specifically, ChatGPT) transparently to write the initial version of the “Artwork title” Wikipedia article. His goal was to better understand the technology and explore its potential applications for Wikipedia editing. The guidelines he developed for this experiment may serve as a model for how experienced Wikipedians might use the technology in a responsible way.
*We truly understand the many serious ethical considerations (e.g. using a tool with a high environmental footprint trained on copyrighted material), but we wish to focus on the implications of using these tools for editing at this time.
Author name(s):
Wikimedia username(s):
- Doctorxgc
Affiliated organization(s):
- Wikimedia NYC; LaGuardia Community College
Estimated length of session
- 60 minutes or more
Will you be presenting remotely?
- I will present in-person
Okay to livestream?
- Livestreaming is okay
Previously presented?
- Teaching Matters Generative AI Series at the City University of New York, Feb. 21, 2025
Special requests: