2025-08-06 –, MOMBASA (🗣️ ar, es, fr, sw)
Language: English
🎥 Session recording: https://youtu.be/Dw9o8Lsl974?t=15910 🎥
This lecture explores the emerging challenge of undeclared AI-generated content in Wikipedia. As large language models such as ChatGPT trained on Wikipedia increasingly contribute content back to the platform, we face critical questions about information integrity. The session examines how this practice undermines Wikipedia's reliability and the various motivations that drive such contributions. Through real-world examples and technical analysis, we'll discuss detection methodologies and policy implications. The presentation aims to foster dialogue about necessary disclosure standards and verification methods to maintain Wikipedia's reliability as a knowledge source in an era where distinguishing between human and machine-written content becomes increasingly difficult.
This lecture investigates the complex relationship between Wikipedia and large language models (LLMs), focusing on the detection and implications of undeclared AI-generated content. Using technical detective work and data analysis, I will reveal surprising findings about how AI-generated text appears in Wikipedia articles without attribution or disclosure.
The session explores a problematic circular information flow: Wikipedia content trains LLMs, which then generate content that enters Wikipedia without attribution, potentially creating a feedback loop of increasingly unreliable information. Through specific examples discovered during my research, I'll demonstrate how this phenomenon extends beyond simple errors to include more sophisticated forms of misinformation.
I will examine the motivations behind incorporating AI-generated content into Wikipedia, from legitimate uses (drafting, summarizing complex topics) to problematic applications (academic shortcuts, ideological manipulation, evading editorial oversight). The lecture will highlight how AI-generated content presents unique challenges compared to traditional forms of misinformation.
The presentation will cover various detection approaches and their limitations, followed by a discussion of policy implications. What disclosure standards should exist for AI contributions? How might Wikipedia adapt its verification processes? What responsibilities do AI developers have in preventing misleading outputs?
Drawing from my experience as a Wikipedia author since 2002, co-founder of Wikimedia Germany in 2004, and former board member and employee of Wikimedia Germany (2009-2014), I'll provide historical context for today's challenges. This perspective allows us to distinguish between perennial issues in collaborative knowledge creation and genuinely new threats posed by undeclared AI content. The lecture will emphasize solutions that honor Wikipedia's core principles while addressing emerging technological realities, avoiding both alarmism and complacency about the integrity of our shared knowledge resources.
- What program track best fits your session? This is your session’s primary thematic category.
-
Community Health
- What other themes or topics does your session fit into? Please choose from the list of tags below.
-
Capacity building, Other
- How does your session relate to the event theme: Wikimania@20: Inclusivity. Impact. Sustainability?
-
Sustainability: How can Wikipedia survive in a world of bots and machine generated text
- What is the experience level needed for the audience for your session?
-
Some experience will be needed
- How do you plan to deliver this session? You will be asked to confirm this closer to the date in case of changes to the format.
-
Onsite in Nairobi
- Should your session be selected for the program, do you agree to release your session and supporting materials on-wiki and on the eventyay platform under CC BY-SA 4.0?
-
I agree
- Etherpad for the session:
I joined Wikipedia as an author in 2002/2003. I co-founded Wikimedia Germany in 2004, I served on the board of Wikimedia Germany for a few years and I worked for Wikimedia Germany as a full time employee from 2009 to 2014.
I worked at the German National Library, the European Parliament, the German federal Parliament, a Berlin-based NGO on human rights and civil liberties and I recently started working in the policy team of a company.
My last Wikimania was London, it feels like ages ago.