The federal government has been spending millions on AI-powered censorship tools through the National Science Foundation (NSF), a recent congressional report revealed.
With an annual budget of almost $10 billion and more than 1,500 federal employees, the NSF funds 2,000 grantees annually and contributes approximately 25% of federal funding for research at US universities. In 2021 and 2022, the NSF doled out about $16 billion in grants.
A report published last week by the House Judiciary Committee and the Select Subcommittee on the Weaponization of the Federal Government revealed that the NSF allocated at least $13 million towards various censorship technologies. These programs, mostly developed by researchers at American universities, use AI algorithms to root out “misinformation” and “disinformation” online.
But while those terms are often used as euphemisms for censorship, documents show that NSF officials were aware that the tools they were funding would be used to suppress speech online.
When the University of Michigan pitched its WiseDex tool to the NSF Convergence Accelerator, for example, it marketed the program as a way for social media platforms to “externaliz[e] the difficult responsibility of censorship.” The NSF granted the university $750,000 for WiseDex.
Co-Insights, a censorship tool developed by the nonprofit organization Meedan, was designed “to counter misinformation online.” As part of this mission, Co-Insights promised to weed out online content “undermining trust in mainstream media.” NSF awarded Meedan $5.75 million.
Researchers at the University of Wisconsin-Madison presented the NSF with CourseCorrect, an AI program specifically centered on “address[ing] two democratic and public health crises facing the U.S.: skepticism regarding the integrity of U.S. elections and hesitancy related to COVID-19 vaccines.” The NSF awarded the university $5.75 million for CourseCorrect .
Similarly, the Massachusetts Institute of Technology (MIT) pitched Search Lit, a tool which also focused on suppressing unfavorable content about elections and the COVID-19 shots.
“In particular, the MIT team believed that conservatives, minorities, and veterans were uniquely incapable of assessing the veracity of content online,” the congressional report said.
MIT was awarded $750,000 for the Search Lit platform.
These were just four tools among 12 presented to the NSF as part of a program launched in March 2021 called Trust & Authenticity in Communication Systems, for which $21 million in taxpayer funds were allocated.
These automated censorship platforms are game-changers for the federal government, which has been using a manual “switchboarding” process to suppress information. In this process, officials from the Cybersecurity and Infrastructure Security Agency’s (CISA) “Mis-, Dis-, and Malinformation” team submit unfavorable content to social media executives who have them scrubbed from the platforms. CISA engaged in this practice for the 2018 and 2020 elections.
During the 2020 election cycle, CISA also partnered with Stanford University, whose Election Integrity Partnership (EIP) program employed around 100 operatives working in shifts to scour the internet for unfavorable posts. The EIP flagged thousands of such dissenting posts which they sent to the social media platforms for removal.
