Future Audiences/Experiment:Add a Fact

Add A Fact is a new experimental feature being developed by the Future Audiences team that builds off the lessons from Citation Needed.

As with all Future Audiences experiments, Add A Fact seeks to contribute to the Wikipedia community’s work while attempting to prove or disprove a hypothesis. In this case, we’re seeking to understand how editing audiences can make editorial contributions off-platform (that is, without going directly to Wikipedia.org), and if generative AI can support or hinder this process.

We hypothesize that:

being able to add facts to Wikipedia in a light-touch way can help editors speed up their process without interrupting their day, and that we can make LLM-in-the-loop tools that support rather than obstruct editors’ work.

Using the Add A Fact Chrome browser extension, an editor can select text from another website that they may want to add to a Wikipedia article, use an LLM to check if the text they selected is relevant to any articles, and whether the article agrees or disagrees in full or in-part with the text. After the user selects an article Add A Fact will then send the text, their thoughts on the text and a structured citation (using Citoid) to the talk page of the article they select.

We recognize how adding facts off-platform to talk pages could overwhelm them and the article watchers who manage those pages, so to that end, Add A Fact users will be limited to sending a maximum of 10 facts per day.The MVP release, slated to be ready by Wikimania 2024, will be limited to autoconfirmed en.wiki editors with accounts. In other words, at first, Add A Fact will not be available to IP editors, or just anyone on the Internet at-large.

Add A Fact is subject to change based on user feedback. You can follow Add A Fact’s development in Phabricator.

Timeline

The team is currently evaluating ways to test the research questions below in the fastest, lowest-cost way – including via existing tools (i.e., the Citation Needed extension) and/or developing new experimental tools or features.

Research questions

1. Do people on the internet want to contribute good-faith information to Wikipedia?
2. Who are the people who would be interested in doing this? i.e.:
   • The general public – people who have a casual relationship to Wikipedia (are aware of and may visit it from time to time, but wouldn't consider themselves members of our movement, may not donate, etc.)
   • People who are Wikipedian-like in some way – e.g., Reddit moderators, subgroups on the Internet (i.e., fandoms, communities, fact-checkers, etc.); donors
     • What could incentivize non-Wikipedians to do this? i.e.:
       • Add extra incentives: i.e., wrap the "add a fact" functionality into another useful end-user tool, e.g. Citation Needed (if we discover it is useful/attractive to end-users)
       • Radically lower the barrier to entry: i.e., make the functionality run in the background, like spellcheck (checking for and identifying claims that look like they are on reliable sources and should be added)
       • Other?
   • Existing Wiki(p/m)edians
3. How might we deliver these contributions into existing or new pipelines for human review/oversight/addition to Wikipedia?

See also

• WikiGrok (2014-15): on-wiki experiment to encourage casual Wikipedia readers to contribute a structured Wikidata fact to a topic (by answering a simple question about the article they were reading).
  • Findings: high overall engagement and quality of responses (especially when aggregated). Main blocker was in the cost to maintaining/scaling the infrastructure to power suggested questions (at the time, a graph database was the best solution, but there were no affordable, scalable open source solutions on the market).
• Citation Hunt: a game hosted on Toolforge that allows anyone to search for/add a reference to an unsourced claim onwiki.
• Wikidata for Web: an extension that displays data from Wikidata on various websites and also allows extraction of data from these websites to input into Wikidata.
• Article Feedback Tool: a tool piloted to engage readers to participate on Wikipedia and to help editors improve articles based on reader feedback.
  • Findings: Readers welcomed the opportunity to engage with Wikipedia in a new way, but since they were asked to provide freeform feedback on article quality, most of their output was not useful to improving the content. From the final report: “Over a million comments were posted during this experiment: on average, 12% of posts were marked as useful, 46% required no action, and 17% were found inappropriate by Wikipedia editors. However, a majority of editors did not find reader comments useful enough to warrant the extra work of moderating this feedback.”