We need to talk Some More about Wikipedia
Earlier this week, I wrote about Wikipedia’s bias problem. Wikipedia has become one of the most popular training sets for AI researchers. Sure, it underpins well-known and widely used services like Alexa and Siri. But more than that, it serves as the massive and free feedstock for countless other products.
Whatever you think about Wikipedia as a service, I argued, it’s in the best interest of our democracy that the service approximates knowledge as broadly as possible. It’s already a stand-in for a host of emerging AI-powered products and services.
It’s impossible to overstate how passionately LinkedIn members reacted to the idea. Many had edited Wikipedia themselves, or donated to help keep it running. Most agreed that for the for the future of AI, it’s important to shore Wikipedia up now.
“Actual AI practitioner here,” wrote Facebook software engineer Michael Hanson. “I can confirm that, yes, Wikipedia is used as feedstock for many, many AI projects, along with numerous other Wikimedia projects that are less visible to most internet users.” Hanson explained why engineers often turn to Wikipedia over other data sets, writing that there “is no other freely available data source” that is large, comprehensive and released under a liberal license so it’s suitable for academic research as well as commercial projects.
“For those commenters who think the Wikipedia is full of errors and lies, and that using it as feedstock for AI projects is madness: we know,” Hanson writes. “ The world is full of errors and lies, both in digital and in print. Spammers, hackers, true believers in any number of delusions, and polemicists skilled in the dark arts of rhetoric are a fact of life for anyone trying to organize the world's information. The fact is, the Wikipedia is better than most of it, and easier to use than all of it.”
Hanson notes that it’s hard to reproduce the energy of a community project that is working, as Wikipedia is. He suggests instead that we agree on measurable metrics for improving the service, and then try to improve them.
That sentiment was echoed by many LinkedIn members, who had myriad creative ideas for how to make Wikipedia more inclusive. Some were extremely practical. Education entrepreneur Christine Renaud is planning to gather people for a French-speaking Edit-A-Thon in Montreal in early 2019. If you’re interested in attending, follow her. Salt Lake City-based entrepreneur Annette Shade is planning to start adding women business leaders and entrepreneurs. “Please send me any business woman you think should be added and why (of course, please include any relevant sources),” she writes. “I’ll add them to my list.”
Insurance executive Paul Ivanovskis points to the work of Jess Wade, who wrote 270 Wikipedia pages about female scientists last year in an effort to get her female peers noticed. “I kind of realized we can only really change things from the inside,” she told The Guardian. “Wikipedia is a really great way to engage people in this mission because the more you read about these sensational women, the more you get so motivated and inspired by their personal stories.”
Several people pointed out they felt Wikipedia’s demands for proving a subject worthy of inclusion were unreasonable. Entrepreneur Gerard McLean says he quit trying to edit articles as a volunteer “because most edits and entries are immediately challenged by editors who have issues with arrogance and hubris.” This sentiment came up several times in comments, with people complaining that their attempts to add entries were turned down over questions of adequate sourcing.
Others took issue with Wikipedia’s current sourcing requirements for its entries. They believe Wikipedia’s criteria for identifying a reliable source is too narrow. Crypto entrepreneur Chet Zeiger pointed out an alternative to Wikipedia, Everipedia, which allows a broader definition of sourcing and incentives contributors through the blockchain.
Comments aren't closed on this topic, and owing to the extreme interest, I will plan to revisit it through my reporting later this winter. If you know of interesting projects designed to fortify Wikipedia, or compelling alternatives to Wikipedia for AI researchers, please be in touch!
Thanks for reading my series. I've reported on tech's business figures for 15 years. I share reporting and reflections like this piece on Wikipedia featuring the Canadian optical physicist Donna Strickland. I hope you'll subscribe.