As conservatives look for new ways to hold academia accountable, some see artificial intelligence as a tool to help examine the legitimacy of academic studies and papers.
Coming off recent incidents in which critics exposed apparent plagiarism from high-profile academic leaders, some Republicans are feeling confident that their ability to scrutinize these institutions has only gotten stronger.
AI may make it easier to detect academic fraud and scale those efforts to land the largest blow to the legitimacy of universities yet.
“This is going to revolutionize academia because so many people have skated by, basically copy-pasting and doing it in a way that may be a little creative but where they’re basically just taking other people’s ideas and reprinting them,” Jon Schweppe, director of policy and government affairs at the American Principles Project, told the Washington Examiner. Schweppe said the phenomenon is true across academia, from hard sciences to humanities.
“Millennials grew up with this idea that if you could just cite a study, whatever you say is bond. I think if that is destroyed and AI is part of it, well, good riddance,” he continued. “Someone without resources, or even without expertise, could theoretically use AI to find weaknesses in the arguments of people who are more qualified or more prestigious or whatever. That is the brilliance here.”
The power of academia and the search for fraud
Critics say academia has become an elite echo chamber led by people who prioritize political advocacy and social experimentation over inquiry. The problem, as they see it, is overwhelming: The vast majority of academics are left-wing, and many academic papers serve as attempts to justify the authors’ underlying politics.
“Being in the transgender fight, this has been kind of egregious to see, but a lot of these studies are industry-funded. These people are not coming in for science. They’re going in to find an outcome,” Schweppe said, adding that many of the academics conducting the studies are simply happy to oblige. “If they try to conduct the study and that’s not the outcome they want, obviously they’re just going to trash the study because that’s not why they’re getting funded. Trying to use AI to help us suss out this stuff, I think that’s going to be a really good use.”
Schweppe was referring to the growing set of evidence showing the “gender-affirming care” model of responding to gender dysphoria is not only ineffective but is dangerous. Nevertheless, certain medical organizations, hospitals, and doctors continue to promote this, sometimes by relying on what critics consider warped research.
Because of the ideological skew in higher education and its culturally prescriptive output, skeptics have searched for ways to rein in the institutional power held by universities and their professors. Some are looking to AI to do the job.
“I think you saw it in the Claudine Gay case that the tools already exist to cause devastation to the current model and really unearth a lot of things that pass the eye test on the first go-around,” Jake Denton, research associate at the Heritage Foundation’s Tech Policy Center, told the Washington Examiner.
Denton was referring to Gay, the ousted former president of Harvard University, who is likely the most well-known recent example of technology’s power to wreak havoc over the institutional power of universities, although it is not clear whether AI or more traditional tools were used to expose her academic history. After critics detected multiple instances of plagiarism in her political science papers earlier this year, Gay resigned, becoming the shortest-serving president in the school’s history.
Researchers have used AI to uncover fraud in the hard sciences, too. A data manipulation expert used AI software earlier this year to help detect plagiarized images across multiple papers written by a top neuroscientist at Harvard Medical School. That neuroscientist, Khalid Shah, was accused of dozens of instances of falsifying scientific data and plagiarizing pictures.
What can AI really do right now?
Plagiarism scandals have given academia hawks new hope and a new sense of innovation.
Those interested in undercutting academia’s cultural stranglehold are now wondering how far the technology can go, asking: What if there were a way to track plagiarism back decades to uncover a long-running chain of academic fraud? What if there was a tool capable of reading data found in a study and invalidating the conclusions drawn by its author? Can artificial intelligence be the answer?
The answer, according to Christopher Alexander, chief analytics officer at Pioneer Development Group and co-founder of Lever AI, is: maybe, but not yet. And don’t be surprised when you’re disappointed.
“What AI has done when it comes to academic plagiarism is it has made it easier to interact with and understand what you’re looking at. Through the natural language processing, you can talk to it and query it on the one hand, but on the other hand, it can process more data and it can find more kind of underlying patterns and detect things than ever before,” Alexander told the Washington Examiner. “Finding mistakes in calculations, finding quotations or things that haven’t been cited, it’s going to be exceptional at that, and it’s only going to get better.”
While AI can be used fairly easily to detect things like plagiarism, it may not be able to detect the deeper forms of fraud many Republicans hope it can expose, such as finding whether the conclusion in an otherwise legitimate paper has been drawn such that the data do not justify it. Alexander said the only way to have AI prove a conclusion wrong is if someone already knows a specific study is incorrect and understands how to direct AI to help find out why.
“AI is exceptional at replicating the work of lower-level, repetitive, white-collar tasks,” Alexander explained. “People confuse the future state that artificial intelligence will likely become with the current state. I think they have an inflated idea of what it is capable of.”
There are several reasons for the limitations of existing AI technology, but the primary one is that AI requires human input and intent in order to function the way a user wants. Its biggest benefit presently lies in its language processing, and nearly all AI tools are simply built off of a handful of foundational models that inform them. One of the biggest benefits of AI is its use of natural language processing, which allows users to ask questions or direct a query in a human language as opposed to through algorithms.
“AIs are not the arbiter of truth, and the designers of it know that. First and foremost, it’s been designed to have a positive user experience. They need you to come back to it, they need you to use it, they want you to rely on it,” Alexander said. “So the AI as a consequence of that, whether intended or not, it’s prima facie duty is to make you happy. It’s not to give you the truth. It’s not to tell you what is right. It is not to do anything more than give you an answer you’re satisfied with.”
The usefulness of AI as an academic fraud detection tool depends on what the user means by “fraud.” It can be directed to find plagiarism, for example, so long as it has a universe of data with which to compare the paper, albeit at a much faster processing speed than the “lower-level … white-collar” worker can, Alexander said. But for more complex types of academic misconduct, users must have advanced knowledge both of the fraud they hope to uncover and of how to program AI to find it.
AI tools, at least right now, are best suited for different stages of the research process, such as grant approval, where officials approving grants may not be as capable of spotting the “nonsense” a researcher will include, Denton said.
“Sifting good from bad is where this really would thrive,” Denton said, comparing its use to an AI-powered job hiring software finding the top applicants out of a pool of thousands. “It is distinguishing between a dud proposal before you go through all the legwork and research to suddenly realize that was a bad proposal.”
“We’re probably not far away from a ‘peer review bot’ or something of that nature. I think that is where you can see a huge enhancement on the quality of these reports,” Denton added. “The average researcher knows they can get away with a ton in their different issue area because the audience isn’t informed so they push the limits of their assumptions, and a ‘peer review bot’ solves for that gap, the knowledge gap between the reviewer, maybe the target audience, and the researcher themselves.”
AI has also been used to facilitate academic fraud, with some authors and researchers using the technology to fabricate papers, or at least do some of the work for them. That problem is compounded through what are known as “paper mills,” which the Committee on Publication Ethics defines as “profit oriented, unofficial, and potentially illegal organizations that produce and sell fraudulent manuscripts that seem to resemble genuine research.”
Last month, Wiley, a 217-year-old academic publisher, was forced to close 19 journals after it was caught in a widespread fraud scandal that saw over 11,300 papers retracted after the journals accepted research from paper mills.
Despite that, publisher Science announced in January that all papers would be required to go through an AI review to detect image fraud, designed to detect whether scientists have manipulated images to support an outcome that is not actually justified by the data.
The black box
One of the biggest problems with using AI to help detect academic fraud is that many programs are infected with information bias. The “black box,” which is the proprietary algorithm powering how the AI runs, can and often is written to include left-wing biases affecting the objectivity of the function.
“The tools in their purest form are raw, unbiased, trying to cut right to the data and present you the factual findings,” Denton said. “You can always add in ‘trust and safety’ layers that will skew this tool, this model, the same way that they’re skewing the dataset as a bias researcher.”
Because most AI tools are simply layers built on top of the few existing foundational tools, the information dataset an AI is using to find answers is often the same and infected by the same biases. In addition, using the current technology, it is also not practical to make brand new tools that are not informed by the few foundational models due to high energy use and costs.
One recent example of these biases at work is Google Gemini’s production of racially diverse Nazis. In February, the AI tool gained widespread attention after users prompted it to depict Nazis and the tool depicted them as being black and other nonwhite races.
“Look at what the thing does — it makes black Nazis! That’s insane. What people don’t realize when you see the output of an AI, that’s how it thinks,” Alexander said. “It is so diversity-crazy that it’s like, ‘Oh my God, all these Nazis are white. That’s not inclusive. I’ve got to fix it.'”
Alexander said that even though the diverse Nazi incident sparked mockery of left-wing Google workers, it speaks to a much more concerning reality about AI tools.
“The AI draws from the internet, so the AI is reading about how important diversity is and the people who are programming it are telling it that it’s really important to be diverse,” Alexander said. “What really concerns me is the amount of nonsensical drivel that has been produced by academia in the last 20-plus years is really guiding how the AI works.”
CLICK HERE TO READ MORE FROM THE WASHINGTON EXAMINER
AI tools right now will actually “actively work against” someone genuinely trying to find new discoveries because a researcher will be “working against the prevailing body of knowledge in any profession” that is informing AI output, Alexander explained.
“I’m actually really worried that it’s going to get worse the way the internet has,” Alexander continued. “It’s just ingesting stupidity all the time, and we already know that the major search engines are cooking the books, suppressing content, the U.S. government is suppressing content, so these things are going to be ridiculously biased.”
Breccan F. Thies is an investigative reporter for the Washington Examiner.
