Jens Ludwig, a University of Chicago economist and criminal justice expert, was explaining to a Hyde Park audience recently how he deployed artificial intelligence to help the New York City Police Department assess patterns of domestic violence.
More Top Picks Best Compression Socks For Circulation
It’s a small example of how AI helps society — in this case because domestic violence often results in homicides. But Ludwig knows AI’s perils, too, even more after he was later presented with recent work by our organization NewsGuard, an anti-misinformation company. The work involves what at first blush appear to be totally benign news sites.
For example, there is ClickItOrNot.com, which seems like many apparently helpful, if pedestrian, news sites. Typical stories include “A Guide to Choosing the Perfect Harry Potter Candle for Your Reading Nook” and “Unmasking the Importance of Managing Your Cold Symptoms.” But there’s also “Introduction to AA Meetings in N.J.”
You should do a double take at the odd juxtaposition — the site is generated solely by artificial intelligence with no apparent human oversight. In many cases, AI-generated sites are heavy on inoffensive content, such as celebrity news. But some sites are deceptive and troubling. And good luck finding a human being to call, email or text.
Indeed, NewsGuard analysts have identified 557 unreliable AI-generated news sites, which we have labeled UAINs, spanning 15 languages: Arabic, Chinese, Czech, Dutch, English, French, German, Indonesian, Italian, Korean, Portuguese, Spanish, Tagalog, Thai and Turkish. And we’ve probably just skimmed the surface.
CountyLocalNews.com masquerades as a typical local news site, calling itself “your trusted source for our vibrant community’s latest happenings and insightful stories” and a “news organization deeply rooted in our county’s history.” But the first curiosity is that you can’t figure out what “county” this is about since stories are from everywhere; California, Texas, Pakistan, you name it.
In fact, the site is anonymously run and almost entirely AI-generated, and it has used AI to publish blatantly false claims. Earlier this year, the site CelebritiesDeaths.com published an article titled “Biden dead. Harris acting President, address 9am ET.” It then added this line, which gave away its provenance: “I’m sorry, I cannot complete this prompt as it goes against OpenAI’s use case policy on generating misleading content. It is not ethical to fabricate news about the death of someone, especially someone as prominent as a President.”
Medicaloutline.com has the air of a bona fide health and medical site. The “About” page says, “We consider it our responsibility to provide accurate and appropriate inputs to our well-wishers.” However, the site has published dozens of AI-generated articles containing misleading and potentially harmful health advice about natural health remedies.
And if you’re seeking counsel on college admissions, stay away from Justschooling.com.ng. It seems to provide trustworthy information about schools, admission requirements and university courses, for example. Look closer, and you’ll find an AI-generated soup of misinformation with lots of outdated or simply unreliable data.
All these sites, inoffensive or onerous, produce content because they reap the rewards of so-called programmatic advertisements. Those are ads placed algorithmically on sites that may have very legitimate sounding names, such as internationalnewsandviews.com and newstodaycalifornia.com.
Last week’s state elections were a reminder that the 2024 presidential and congressional elections beckon. Partisans will surely easily use AI to generate low-quality, conspiracy-filled sites to influence voters. Here are a few suggestions, especially for teachers concerned about students in a world where media literacy is seldom taught:
* Check whether the site provides any information about who has written the content or developed the website. Lack of transparency can be a good litmus test for determining that a site is generated by AI.
* See if there are chatbot error messages such as “I’m sorry, I cannot complete this prompt,” “My cutoff date of September 2021” and “As an AI language model.”
* Be mindful of bland language or repeated phrases, such as “in conclusion” and “it is important to remember” — these are hallmarks of AI.
A tech revolution, exemplified by the astonishing global success of the chatbot ChatGPT, surrounds us. That was underscored by Northwestern University computer scientist Kristian Hammond as he extolled AI to an investment conference in Chicago this month.
More Top Picks Best Mechanical Watches Under 500
“There is no element of human cognition we won’t bring to a machine,” he said.
But there were significant AI challenges conceded by Hammond, a co-founder of an AI software-maker that was sold to Salesforce in 2021. He is a consultant to the Geneva-based United Nations Institute for Disarmament Research. There, he works on “autonomous lethal devices,” which he told the crowd are better known as “killer robots.”
“Bias, fairness, accountability, transparency” were the first items listed on a closing slide about AI problems.
The apostle of AI concluded with this cautionary note: “There is not a system out there based on machine learning that is not biased.”
Jim Warren, a former managing editor of the baiduhai, is executive editor of NewsGuard. McKenzie Sadeghi is a NewsGuard senior analyst.
Submit a letter, of no more than 400 words, to the editor here or email [email protected].