WIKIPEDIA IS BIASED AGAINST CONSERVATIVES AND THE SLANT IS INFFECTING AI MODELS

National Review

 

Wikipedia Is Biased against Conservatives — and the Slant Is Infecting AI Models

By JAMES LYNCH

June 20, 2024 8:02 AM

 

A new study released on Thursday by a conservative think-tank is giving scholarly credibility to long-held conservative suspicions of bias among Wikipedia editors on entries related to current events.

 

Wikipedia entries for conservative political figures and organizations do in fact contain more negative attitudes than entries for their liberal counterparts, according to a new Manhattan Institute report released Thursday. This bias could have profound implications for the training of large-language artificial intelligence models, according to the study’s author, David Rozado, a computer scientist who previously researched the apparent left-wing bias of artificial intelligence chatbot ChatGPT and other large-language models.

 

“In general, we find that Wikipedia articles tend to associate right-of-center public figures with somewhat more negative sentiment than left-of-center public figures; this trend can be seen in mentions of U.S. presidents, Supreme Court justices, congressmembers, state governors, leaders of Western countries, and prominent U.S.-based journalists and media organizations,” Rozado’s report states.

 

“We also find prevailing associations of negative emotions (e.g., anger and disgust) with right-leaning public figures and positive emotions (e.g., joy) with left-leaning public figures,” the report adds.

 

The terms ‘Donald Trump’ and ‘President Trump’ have the most negative sentiment attached to them of recent presidents. “Barack Obama,” is the most positive, followed by ‘Jimmy Carter’ and ‘President Biden,’ the study found.

 

Most of the entries for senators contain positive sentiments, though entries for roughly a dozen Republicans and just one Democrat contain negative sentiments.

 

Among the House lawmakers, a noticeable portion of the entries for Republicans and just a couple of entries for Democrats contain negative sentiments. As the data goes in a more positive direction, the red fades out and blue dominates the House sentiment chart, reflecting left-wing bias.

 

The political bias is similarly noticeable with recent Supreme Court justices. The terms “Brett Kavanaugh,” “Amy Coney Barrett,”  and “Justice Alito” are all associated with negative sentiments.The entry for deceased liberal justice Ruth Bader Ginsberg has overwhelmingly positive sentiment, the most of any recent justice. Progressive justices Ketanji Brown Jackson and Sonia Sotomayor have the second and third highest positive sentiments, respectively.

 

Rozado’s study also found a similar bias in entries for U.S. media organizations and think tanks, with liberal organizations receiving more favorable treatment.

 

Some of the political associations embedded in Wikipedia by its millions of volunteer editors have already emerged in artificial-intelligence models developed by OpenAI, according to the report.

 

“Our results suggest that Wikipedia is not living up to its stated neutral–point–of–view policy. This is concerning because we find evidence of some of Wikipedia’s prevailing sentiment associations for politically aligned public figures also popping up in OpenAI’s language models, which suggests that the political bias that we identify on the site may be percolating into widely used AI systems,” the report warns.

 

Wikipedia’s neutral point of view policy is meant to ensure articles are “written with a tone that provides an unbiased, accurate, and proportionate representation of all positions included in the article,” according to the website’s own entry on the topic.

 

Rozado’s study utilized large-language models to assess the sentiment and emotional tone tied to political terms within Wikipedia articles.

 

“The goal of this report is to foster awareness and encourage a reevaluation of content standards and policies to safeguard the integrity of the information on Wikipedia being consumed by both human readers and AI systems,” Rozado concludes.

 

In the intro, Rozado’s report cites a 2012 paper that analyzed 20,000 English-language Wikipedia entries and observed a pro-Democratic party slant. Since 2012, an abundance of opinion polls and academic research have demonstrated a significant increase in American political partisanship, especially with regards to former president Donald Trump.

 

An example of Wikipedia’s apparent political bias came up when the site rapidly changed its entry on the colonial “Appeal to Heaven” flag after Democrats and their allies in the media attacked the credibility of Justice Samuel Alito because his wife flew the flag outside their beach house as a supposed symbol of the January 6 Capitol riot.

 

The flag has been a symbol of freedom since the American founding, and was flown outside San Francisco’s city hall until it was taken down last month when the Alito story became fodder for a prolonged news cycle.

 

Another example of Wikipedia’s apparent bias took place two years ago when the Biden administration disputed the academic definition of recession to defend its economic record, and Wikipedia edited its page to claim the definition of recession was not universally agreed upon. The site temporarily suspended edits to its page when the recession entry became battlefield terrain for prospective editors.

 

Wikipedia’s bias received significant attention earlier this year when NPR hired former Wikimedia Foundation executive director Katherine Maher to be its new CEO. The Wikimedia Foundation is a nonprofit that owns Wikipedia.

 

Maher’s social-media history indicates that she has a long history of left-wing activism and strongly supported Joe Biden’s 2020 presidential campaign.

 

Before Maher’s history came under scrutiny, Wikipedia co-founder Larry Sanger repeatedly warned of what he perceived as the website’s increasingly left-wing bias.