Much of intelligent design theory concerns distinguishing intentional objects from works of chance or natural law. Within the set of unquestionably designed objects, though, questions about intention can still arise. Art historians might question the authenticity of paintings by looking for clues of tampering or copying. Teachers might suspect plagiarism in a student’s term paper. And now, journal editors are dealing with very serious questions of malicious design in scientific papers. In situations never faced by editors before the computer age, they are finding examples of fake papers generated by artificial intelligence programs. Some are so good they fool even the peer reviewers.
Because forensic science is an example of intelligent design in action, ID inference tools must be summoned for assistance in this endeavor. William Dembski illustrated using ID principles within the class of designed objects in his recounting of the case of county clerk Nicholas Caputo, who was caught stacking the deck for his favored candidates by listing them at the top of the ballot more often than could be accounted for by chance. Today’s examples are even tougher to crack. Often there is no element of chance or natural law to eliminate. The research paper, the perpetrator, and the AI software are all intelligently designed. What then?
A news feature in Nature speaks to the growing problem of fabricated research papers. Reporter Holly Else gives hope that there may be ways to detect malicious design: “strange turns of phrase may indicate foul play in science,” she begins.
In April 2021, a series of strange phrases in journal articles piqued the interest of a group of computer scientists. The researchers could not understand why researchers would use the terms ‘counterfeit consciousness’, ‘profound neural organization’ and ‘colossal information’ in place of the more widely recognized terms ‘artificial intelligence’, ‘deep neural network’ and ‘big data’.
Further investigation revealed that these strange terms — which they dub “tortured phrases” — are probably the result of automated translation or software that attempts to disguise plagiarism. [Emphasis added.]
Because AI cannot yet mimic the cultural nuances known by human writers, an AI algorithm programmed to replace words with synonyms can make a cultural faux pas. It has no problem replacing “cloud computing” with “haze figuring.” A human editor can detect the humor in a paper that speaks of “signal to noise” as “flag to commotion.” This lack of nuance in software can help integrity sleuths for now — until AI catches up.
Back in March in Nature, Matthew Hutson warned about “Robo-writers: the rise and risks of language-generating AI.” The latest iteration of a demonstration program called GPT astonished onlookers:
In June 2020, a new and powerful artificial intelligence (AI) began dazzling technologists in Silicon Valley. Called GPT-3 and created by the research firm OpenAI in San Francisco, California, it was the latest and most powerful in a series of ‘large language models’: AIs that generate fluent streams of text after imbibing billions of words from books, articles and websites. GPT-3 had been trained on around 200 billion words, at an estimated cost of tens of millions of dollars.
It was funny until journal editors started seeing GPT-3 generated papers passing peer review. A worrisome graph in Hutson’s article shows exponential growth in artificial neural network parameters since 2018. What will GPT-n be capable of when its computer connections equal or surpass the neural connections in the human brain? The whimsical robot Data (played by human actor Brent Spiner) in Star Trek: The Next Generation, seems prescient. Data keeps striving to be a capable human mimic while struggling to understand human emotions. With more input from billions of words, GPT-n might be able to laugh at its own previous missteps, like howling at its substitution of “colossal information” for “big data.”
AI professor Robert J. Marks (Mind Matters) and neuroscientist Michael Egnor (Evolution News) assure us that AI will never gain consciousness or self-awareness. That gives us philosophical hope that Data will never quite “get there” as a plausible human mimic. But Hutson warns that nothing stops GPT-3 from generating misinformation, hate speech and terrorist propaganda. Garbage in, garbage out. The rise of large language models such as GPT-3 confronts ID theoreticians with two new challenges: keeping ahead of rapid changes in technology (an “arms race”) and distinguishing integrity from intelligence when everything being investigated is intelligently designed.
Integrity Sleuths and “Tortured Phrases”
Holly Else describes how integrity sleuths looking for “tortured phrases” in other journals. One particular journal found other clues to fakery:
To dig deeper, the group downloaded all papers published in Microprocessors and Microsystems between 2018 and 2021, a time frame they chose because an upgraded version of GPT was released in 2019. They identified around 500 “questionable articles” based on various factors. Their analysis revealed that papers published after February 2021 had an acceptance time that was five times shorter, on average, than those published before that date. A high proportion of these papers came from authors in China. And a subset of papers had identical submission, revision and acceptance dates, the majority of which appeared in special issues of the journal. This is suspicious, the authors say. Unlike standard issues, overseen by the editor-in-chief, special issues are usually proposed and overseen by a guest editor, and focus on a specific area of research.
But then, she continues, the sleuths found themselves facing an accelerating arms race:
Microprocessors and Microsystems was not the only affected title — the researchers also found evidence of tortured phrases in papers published in hundreds of other journals. “Preliminary probes show that several thousands of papers with tortured phrases are indexed in major databases,” they write, adding that “other tortured phrases related to the concepts of other scientific fields are yet to be exposed”.
The bug swarms in Starship Troopers come to mind, or the multiplying fighters in The Matrix that keep on coming no matter how hard the protagonists fight. Without better tools for rapidly sifting quality from quantity, the legitimacy of scientific literature is at risk. “It harms science,” said one sleuth who found 860 instances of tortured phrases in a citation database. Another feels that the count of fabricated papers so far detected is probably the tip of the iceberg. Jennifer Byrne from the University of Sydney commented,
“These papers were also found because they were of very poor quality, but there could be more plausible AI-generated papers within the literature that are harder to detect,” she adds.
Other Problems in Undisputed Designs
Sifting the good from the bad has long been a challenge in scientific literature. May N. Berenbaum, the new editor in chief at PNAS, was horrified at the number zombie papers — retractions still being cited — in circulation. In her editorial August 10, she recounted examples of falsified publications still lurking, going back to the early days of the Royal Society. And she can’t do much about it. Preventing publication would amount to prior restraint (even if that didn’t require omniscience). Purging bad publications would raise charges of censorship. Bad papers have value; historians need to see the bad and the good to accurately portray the bumbling path of science. Tracing the changes in subsequent editions of Darwin’s Origin, for instance, helps historians infer his responses to critics. They need to be preserved intact, unmodified by modern editors. Cleaning up shop would also risk erasing “Sleeping Beauties” that might be out there — forgotten papers that might awaken to new, productive research life.
In short, the corpus of scientific literature is a mess. Fabrication is a mess in every human endeavor, of course: in politics, where disinformation campaigns mislead voters; in marketing of fake Rolex watches and fake diamond rings; in the arts, where plagiarism robs content creators of their due. Where are big lies and half truths not a problem? What’s new in our digital age is the ability for AI to fabricate realistic content and rapidly morph its strategies.
ID Superheroes Needed
The opposite of fakery is integrity. Holly Else uses that word three times, referring to experts who sniff out examples of plagiarism, fabrication and cheating. They are overwhelmed at present. Is this a field where ID advocates can take the lead? Intelligent design theory already encompasses detection of intentionality, whether nefarious or benign. ID researchers would most likely agree that it’s better to have a mistake-ridden paper honestly trying to argue for Darwinism than a made-up paper by a plagiarist or computer program arguing for design.
Considering the warnings above by Holly Else, Matthew Hutson, May Berenbaum and the people they quote, the word of the day is Integrity. The ID filter is great at sorting design from chance and natural law. Now the ID community needs new rules to sift integrity from fakery in online content.
A thought experiment shows how materialism is ill-equipped to tackle this challenge. Imagine journal editors outsourcing the job of integrity-sleuthing to AI programs. They might succeed at the level of counting instances they were programmed to detect, but who watches the watchers? Say another AI program watches the integrity-sleuthing AI programs. The same question continues up an infinite series.
At some level, real flesh-and-blood human beings who value integrity must be involved. A natural follow-up question ensues: how did natural selection produce integrity in a human mind? If Darwinism only values survival, integrity is a phantom. It is merely a ghost that may have temporal value in one culture or time but no ontological existence. A Darwinian worldview cannot argue that integrity is intrinsically good — worth dying for — no matter what the consensus thinks. ID advocates, by contrast, have the grounds for valuing integrity: it is an eternal value impressed on each human soul by intentional design. They alone, therefore, can approach the challenges of sifting the good from the bad in scientific content without jumping ship if the going gets tough. Let ID advocates rise to superhero status to help solve this growing crisis.