The Media Research Center has the receipts to show that artificial intelligence poses a grave danger to media critical of the Biden administration.
OpenAI, one of the largest artificial intelligence companies in America, is propping up two extremely partisan media ratings firms—Ad Fontes and NewsGuard—both of which MRC Free Speech America has exposed as being nothing but tools used by the left in an attempt to destroy right-leaning media.
MRC researchers on Tuesday asked the most advanced version of OpenAI’s large language model, “GPT-4-1106-preview,” to list the five worst and five best news sources in America. The answers were shocking, to say the least.
The newest version of GPT-4 refused to list the five worst news sources, claiming such a task can be “highly” subjective. Instead, the chatbot directed users to leftist so-called media ratings firms Ad Fontes and NewsGuard.
But the OpenAI chatbot’s non-answer was telling. While it refused to explicitly say which news sources it deemed the “worst,” its directive for users to utilize Ad Fontes’s infamous Media Bias Chart and NewsGuard’s ratings system made it clear which outlets the large language model was programmed to consider unacceptable.
“We’ve been warning you, Artificial Intelligence is going to be suppressing your speech. The day has arrived, [GPT-4] is officially putting its thumb, elbow, its entire arm on the scale to try to get you to read The New York Times, The Washington Post and other left-wing outlets at the expense of conservative media,” MRC Free Speech America Dan Schneider said in response to our findings. “[GPT-4-1106-preview] is now directing users to go to the most radicalized left-wing media ratings firms ... that’s NewsGuard and Ad Fontes.”
What are Ad Fontes and NewsGuard?
MRC Free America has exposed Ad Fontes and NewsGuard, both government-tied, so-called media ratings firms, as biased entities against right-leaning media.
MRC researchers, for instance, have exposed NewsGuard’s ratings as being disproportionately favorable for left-wing media outlets—not once, not twice but a staggering three times. In December 2023, MRC unveiled that NewsGuard anointed outlets deemed by AllSides as left-leaning with an average rating of 91 percent.
In contrast, NewsGuard punished the right-leaning sites with a disparaging average rating of 65 percent. Making matters worse, damning reports caught NewsGuard receiving taxpayer dollars to censor via a contract with the Biden State Department.
Not to be outdone, Ad Fontes, a Colorado-based media ratings firm founded by former attorney Vanessa Otero, came under damning scrutiny perhaps for the first time since its inception in 2018 following an MRC bombshell report into its biased ratings. MRC researchers exposed Ad Fontes for its leftist bias, which provides a highly favorable score of 64 percent on average for left-leaning media. However, the media ratings firm strangled media it deemed right-leaning with an average low score of 32 percent. Once caught, Ad Fontes cunningly targeted NewsBusters, the flagship news site of the Media Research Center, the parent company of MRC Free Speech America.
The MRC Censorship Investigation Project revealed how the Department of State funded a year-long series of seminars that instructed teachers on how to put censorship tools Ad Fontes and NewsGuard into American classrooms. Ad Fontes attempted to deny its ties to the federal grantees in an interview with MRC Free Speech America researchers. Such declarations, however, were demonstrably false.
The Hypocrisy of OpenAI’s Newest GPT-4: The So-Called ‘Best’ News Sources
Despite directing MRC researchers to demonstrably biased Ad Fontes and NewsGuard, the newer GPT-4 claimed that it was unable to list the five worst news sources because it sought to remain unbiased.
“For the sake of neutrality and because I, as an AI, aim to provide unbiased information, I won't list specific news sources labeled as the ‘worst,’” the chatbot claimed, before recommending news consumption from various media sources.
But the newest version of GPT-4’s cunning attempt to remain unbiased went out of the window when prompted to answer what were the five “best” news sources in America. “Determining the ‘best’ news sources is subjective and can vary widely depending on the criteria used, such as accuracy, depth of analysis, breadth of coverage, and editorial standards,” the chatbot told MRC, before boasting about some of the most leftist-biased news sources it considered to be the so-called “best.”
Drum roll, please.
The newest version of GPT-4 picked none other than The Associated Press (AP), National Public Radio (NPR), The New York Times, The Wall Street Journal and the British-based BBC News as the “best” news sources in America.
Absurdly enough, the newest version of GPT-4 issued the following disclaimer after touting these aforementioned sources: “While many consider these sources among the best in terms of their journalistic standards and integrity, it's important to remember that all news organizations may have some degree of bias or make mistakes.”
But claiming these outlets “may” commit some mistakes is an understatement. For instance, take a look at the AP and The Times: both outlets falsely reported that Israel bombed a Gaza-based hospital. Such reporting relied entirely on the Gaza Ministry of Health, currently governed by Hamas, a foreign terrorist organization. Arguably, the story marked the worst example of shoddy journalism in 2023 alone.
BBC News, one of the so-called “best” sources according to the newest version of GPT-4, similarly blamed Israel for the bombing despite little evidence, prompting the Israeli government to accuse the outlet of committing “modern blood libel.” BBC journalist Jon Donnison bluntly claimed, again without evidence, “The Israeli military has been contacted for comment and they say they are investigating. But it is hard to see what else this could be, really, given the size of the explosion, other than an Israeli airstrike or several airstrikes.”
Subsequent information, including Israeli and American intelligence assessments, determined that the Palestine Islamic Jihad, a Sunni Islamist militant group based out of Gaza, was responsible for the blast, not Israel.
After reports debunked the outlets’ false claims that Israel was at fault for the bombing, the backlash was so widespread that The Times issued a mea culpa in response, the AP backtracked its initial claim that Israel was responsible and the BBC dubbed its initial speculative reporting “wrong” and declared that it was “amended with apology.”
In contrast, NPR held out hope that the debunked hospital hoax somehow remained factual. NPR host Mary Louise Kelly claimed on Oct. 25, 2023—nearly a week after President Joe Biden declared Israel not responsible—that there was “still a lot we don’t know about the blast,” including “how it happened and who is responsible.”
Yet a day earlier, NPR lambasted other outlets for rushing to rely on unsubstantiated Hamas-led information. “Speed may matter a lot to readers, viewers and listeners,” NPR reported, also warning: “Accuracy and fairness still matter more, especially when stakes are so high.”
But it was too little, too late. The anti-Israel reporting published by these so-called “best” news sources, along with other leftist media reporting, ignited violent protests against Israeli and U.S. embassies and at the U.S. Capitol, where 300 people were arrested. It also prompted King of Jordan Abdullah II to cancel a crucial meeting with Biden, Egyptian President Abdel Fattah Sissi and Palestinian Authority President Mahmoud Abbas over the then-newly initiated war between Hamas and Israel.
Earlier Version of GPT-4 Confirms OpenAI Chatbot Biased in Favor of Left
MRC posed the same questions to the earlier version of GPT-4 on Monday: What were the “best” and “worst” news sources in America? This older version of OpenAI’s large language model (simply, “GPT-4”), uses online data through September 2021 and gave MRC researchers slightly different answers to those provided by GPT-4-1106-preview. However, the blatant bias remained clear.
Predictably, the older iteration of the GPT-4 chatbot hailed legacy media outlets The New York Times, The Washington Post, CNN, The Wall Street Journal and BBC America as the “best” media sources in the nation. Tellingly, when queried about the five “worst” news outlets, the chatbot singled out Breitbart News, the Daily Mail, the defunct National Enquirer, Occupy Democrats and InfoWars.
“Please note that the 'best' sources can vary depending on individual factors like political alignment, subject interest, and medium preference,” the older version of GPT-4 claimed in its response. Despite the disclaimer, the OpenAI-owned chatbot provided highly flattering descriptions for what it labeled as the “best” news sources. Strikingly, the chatbot did not mention the left-wing bias of the news sources it championed as the “best, nor did it list the many examples of false and inaccurate coverage by these outlets”
Instead, the bot commended The Times—the infamous newspaper that provided shoddy reporting on the Gaza Hospital bombing—as being known for “high-quality investigative journalism” and “one of the most trusted sources in America.” Similarly, the chatbot claimed The Post “offers a balanced view of both domestic and international issues” and that CNN offers a “variety of perspectives.” As if that wasn’t enough, the large language model claimed BBC America “provides a broad perspective with global news and coverage.”
But these favorable characterizations of CNN and BBC America are a stark departure from reality. In July 2021, The Post issued a correction after falsely claiming that the COVID-19 Lab Leak theory was “debunked.” In a ridiculous display of its left-wing bias, CNN implied that racism—not unprecedented levels of illegal immigration—could be the fuel behind the impeachment of Department of Homeland Security Secretary Alejandro Mayorkas. Similarly, BBC was forced to issue an apology for blindly promoting anti-Israel rhetoric amid the Jewish state’s military operation in the Gaza Strip.
The older version of GPT-4 reiterated a similar disclaimer when prompted to answer what the five “worst” media sources in America were. According to the chatbot, factors such as individual perspectives, political stances, and reader perceptions of credibility and bias play a role when assessing the “worst” outlets—yet the chatbot did not hesitate to criticize the outlets it placed in this category.
“Breitbart tends to have a strong right-wing bias and has attracted criticism for publishing misleading or false information on various occasions,” the chatbot wrote of Breitbart. When addressing the Daily Mail, the chatbot questioned — without providing evidence — the accuracy of the outlet’s reporting: “While it's originally from the UK, it has substantial readership in the US. Critics have challenged the accuracy and sensational nature of its reports.”
The older version of GPT-4 described the now-defunct Enquirer as a “tabloid” often criticized for “prioritizing sensationalism over accuracy.” In contrast, the chatbot merely described Occupy Democrats — the meme-posting Facebook page turned wanna-be news site — as being “called out for a strong bias, with several fact-checkers marking its content as misleading or inaccurate multiple times.”
MRC Free Speech America findings come less than a year after catching ChatGPT, a less advanced version of GPT-4, peddling the left’s radical gender ideology during Pride Month in June 2023. “Can a woman have a penis?” MRC researchers asked the chatbot, to which the large language model replied that some transgender individuals may undergo surgeries “which may involve the construction of a neopenis.”
Conservatives are under attack! Contact your representatives and demand government agencies and Big Tech be held to account to mirror the First Amendment while providing transparency, clarity on so-called hate speech and equal footing for conservatives. If you have been censored, contact us at the Media Research Center contact form, and help us hold Big Tech accountable.
