Every month, the number of articles on Wikipedia grows by 17,000.
In the 22 years since its founding, Wikipedia has represented the paragon of an open, collaborative, neutral, and participatory internet where any human can freely contribute and freely consume an ever-growing body of human knowledge. But that could all change in the age of AI.
//
Wikipedia is the 7th most visited website in the world.
//
//Everyone needs Wikipedia
Artificial intelligence algorithms are deeply dependent on Wikipedia for its high-quality data.
Wikipedia is likely the most important single source for training AI models. It makes up 3-5% of all the data a large language model (LLM) like OpenAI’s ChatGPT uses to train.
An AI algorithm’s coherence and accuracy depend on quality data, and Wikipedia is a trove of (mostly) verified, fact-checked, quality data. Jesse Dodge, a computer scientist at the Allen Institute for AI told The New York Times, “Wikipedia going forward will forever be super valuable because it’s one of the largest well-curated data sets out there.”
Even before the emergence of AI, major tech platforms relied on Wikipedia (and still do):
A 2017 study by three academic computer scientists found that the quality of Google’s search was significantly reduced if results from Wikipedia were removed. The findings suggested that Wikipedia makes Google a “significantly better” search engine.
//The paradox of reuse
There have been concerns for a while that the interdependence between Wikipedia and Google search has negatively impacted Wikipedia, and this could only get worse with AI.
The 2017 study referenced above found that Wikipedia gets most of its traffic from Google, but as Google provides brief summaries of longer Wikipedia articles, Wikipedia’s editors are concerned that it could lead to reduced traffic on the actual site. Why click on the full Wikipedia article when you can read a brief synopsis provided by Google?
The researchers found that declining Wikipedia traffic could lead to fewer donations and fewer contributions (which is how Wikipedia sustains itself). They dubbed this phenomenon “the paradox of reuse”: as Wikipedia becomes more disseminated throughout the wider web via search (and now through AI), the weaker it could become as an institution.
//The AI threat
AI poses even greater threats to Wikipedia and the integrity of digital information:
The paradox of reuse: The paradox of reuse documented with search engines could also happen with AI algorithms. If people prefer to use an interactive AI chatbot to get answers over reading through a 10,000-word article (with 400 footnotes), then page views, donations, and contributions to Wikipedia could wane, imperiling the future sustainability and health of the site.
AI can hallucinate: In contrast to the slow, hard work of fact-checking millions of articles on Wikipedia, AI chatbots are known to hallucinate and stretch the truth. More concerning, though, is how AI might oversimplify complex issues or neglect important information (from historical events to medical diagnoses).
Letting AI algorithms train on Wikipedia: While AI algorithms have already trained on Wikipedia’s high-quality set of articles, there’s growing push-back on the fairness of letting for-profit companies like Google and OpenAI profit off of the free knowledge that’s been carefully curated by a nonprofit. Similarly, many have argued that because AI is dependent on user data, users should be paid for it.
Using AI to write Wikipedia articles: The community of Wikipedians is split about whether to use AI algorithms to generate new Wikipedia articles. There’s concern that machine-generated content could be less factually accurate, dilute the accuracy of Wikipedia, and require significant human effort to fact-check. Wikipedia’s current draft policy suggests that contributors should avoid using AI algorithms, in part due to their implicit biases and the legal exposure of libel suits and copyright violations. If algorithms generated biased on untrue Wikipedia articles, and then algorithms trained on those articles, we could find ourselves in a negative spiral of more and more misinformation.
The challenges Wikipedia face in the age of AI are immense, but they’re not insurmountable. Over the years, the demise of Wikipedia has been predicted many times (in 2002, shortly after it was launched, critics thought of Wikipedia as a prank or a joke). Yet, Wikipedia has adapted and continued to remain one of the most highly-trafficked and important websites in the world.
//The next chapter for Wikipedia
Wikipedia is launching various experiments to adapt and flourish in the age of AI.
New APIs: Because most AI algorithms like ChatGPT aren’t displaying the most up-to-date information (the end of GPT-4’s training period was September 2021), Wikipedia has created a ChatGPT plugin that allows ChatGPT to search for and summarize the most up-to-date information.
Leveraging AI: While the global community of Wikipedians is split on how to leverage AI in the expansion and curation of Wikipedia articles, Leila Zia, the head of research at the Wikimedia Foundation (parent organization to Wikipedia) is developing step-by-step chatbots to help volunteer editors and contributors with best practices for new articles. AI could also ensure a neutral point of view and predict whether a new article or edit would be overruled.
Getting compensated: Because big tech firms and AI algorithms are so reliant on AI, there’s the possibility that Wikipedia could seek compensation from private companies to use its free data. Technically, an AI chatbot that uses Wikipedia content but doesn’t cite the source (as many chatbots neglect to do) violates Wikipedia’s terms of use. The Wikimedia community is considering if any legal recourse exists, or if the community could go on a “data strike” and prevent AI firms from scraping Wikipedia’s data until they pay up (there is precedent for compensation from big tech, with Google paying Wikipedia through Wikipedia's Enterprise division).
//Easy answers vs. hard truths
Wikipedia’s resilience over the years suggests that it will find a way to remain central to the modern internet for years to come. But perhaps it’s not Wikipedia we should be worried about. It’s ourselves.
In an age of AI, where AI algorithms hallucinate, where we’re flooded with more and more AI-generated content, and where citations and footnotes are absent, we will need better skills to distinguish between what’s true from what’s not.
It’s certainly easier to ask ChatGPT for the answer than it is to pore through section after section of a Wikipedia article, checking footnotes and verifying sources. But what do we lose—as a society—when the convenience of getting the easy answer atrophies our ability to do the hard work of finding the truth?
Other notable headlines
//💸 You are probably oversharing on Venmo. An article in The New York Times explored the privacy concerns on Venmo and how to protect your data.
//🖥 An article in Rolling Stone profiled the women who are identifying the biases hidden within algorithms and the lack of representation in the tech companies building them.
//📹 Two apps in India, designed for home security, are being used to watch and harass domestic workers, according to an article in Rest of the world.
//📈 The AI hype that’s driving up Nvidia’s stock price is triggering flashbacks to the dot-com bubble (and eventual bust), according to an article in The Wall Street Journal.
//🤖 An article in The Guardian argued for a cross-sectoral, human-centered approach to shaping artificial intelligence. For AI to be a force for good, it’s going to require participation from everyone.
//🏫 A new social network from Stanford could be the next Facebook on college campuses, but with a Gen-Z twist. Fast Company profiled Fizz, which is already in 80 universities.
//👤 An article in WIRED explored the implications of Zoom’s updated terms of service, what it means for user privacy, and the race to gather data by companies looking to compete in the AI age.
//📱 Social media is provoking greater violence in communities across the United States. An article in The Atlantic explored how feuds that start on Instagram or Snapchat rapidly translate into violence in real life.
Partner news & opportunities
// Virtual event on the history, theory, and practice of digitizing libraries
August 24th at 1pm ET
The Internet Archive is hosting an online webinar discussion to reflect on the efforts to digitize libraries. Join the upcoming discussion with author Oya Rieger, co-author of "Moving Theory Into Practice." This event offers a unique opportunity to delve into the world of digitization and its practical applications. Learn more and register here.
// Independent Evaluation of Meta’s report on the 2020 election
Tom Cunningham, a resident fellow at the Integrity Institute, has conducted an independent evaluation of the recently released 2020 election experiments by Meta. His insights contribute to the ongoing discourse around the role of social media platforms in elections and the broader social internet landscape. Learn more and check it out here.
/ Project Liberty is a advancing responsible development of the internet, designed and governed for the common good. /