They call it AI pollution. As content created by large language models (LLMs) seeps into every domain like an oil leak, it can have two outcomes.
First, it will become more difficult to distinguish between true and false, which will accelerate the falsification or degradation of the information provided by the Internet.
The second result will be a clearing of the sky, somewhat, as a walled garden of protected content, open to users in paid, freemium and other subscription models.
News website paywalls and YouTube Premium are early examples. But across large swathes of the Internet, pollution flows freely.
“Welcome to the world’s first website generated entirely by artificial intelligence!” reads the disclaimer on the Enlightened Mindset landing page.
The site provides no ownership details or contact information, but lists hundreds of articles by “contributors” such as “Happy Sharer” and “Crazy Lee.” Titles include “Is Hagrid Played by a Robot? An Exploration of the Possibilities,” “How to Be More Lady: Speak Softly, Dress Well, and Move Elegently,” and “How to Be Dangerous: Learn to Defend Yourself, Carry a Concealed Weapon, Build a social network”.
The stories contain fictional quotes from real people. Hagrid’s Story includes a story by “Dr Reuben Binns, Professor of Artificial Intelligence at Cambridge University” about the shortcomings of using a robot to play Hagrid. In the real world, Binns is an associate professor in human-centered computing at the University of Oxford.
The number of platforms using this type of approach is growing, and most do not provide a disclaimer.
lie down and wait
Earlier this month, NewsGuard, which tracks misinformation by assigning trust scores and ratings to online news sources, discovered more than 700 AI-generated news and information sites that operate with little human oversight and no disclosures the nature of its content.
New sites are being steadily added to the list, a valiant ongoing effort, but likely just a facade. When NewGuard launched its AI tracking center in May, a team of content analysts worked to identify and flag such content, identifying 49 such sites in the first month.
These platforms publish fabricated reports about political leaders and spread false news about celebrity deaths. In November, the website GlobalVillageSpace.com announced that Israeli Prime Minister Benjamin Netanyahu’s psychiatrist had committed suicide (the article was later marked as satire).
In April of this year, news platforms around the world reported the death of a Canadian actor named Saint Von Colucci, who reportedly underwent multiple plastic surgeries to look like Jimin from the Korean pop group BTS. for complications. Colucci does not exist; all his images are artificially generated.
Al Jazeera called it “the first known case of artificial intelligence being used to deceive the media… heralding a new era of computer-generated fake news.”
Tech industry analyst Kashyap Kompella said that as the number of such sites surges, reliable sources such as trusted media brands will scramble to form digital forensics teams trained to combat their creation. error message. “AI-powered fact-checking tools will also be built into future browsers. This will be the good vs. evil battle trope we see so often in science fiction, with AI on both sides of the battle.”
Zombie websites, ghost products
Take a closer look at these domain names. The old people are slowly coming back to life. The Hairpin, a women’s website that shut down in 2018, has recently been revived and filled with AI-generated content. Headlines here include ridiculous stuff like “Celebrities have tiny real teeth under their big dentures.”
This version of Hairpin is run by Serbian DJ-turned-entrepreneur Nebojša Vujinović Vujo, who recycled some abandoned news domain names and flooded them with LL.M.-generated headline bait.
“In many cases, the revenue model for these sites is programmatic advertising, which the ad tech industry serves without regard to the nature or quality of the site,” NewsGuard said in a statement. “Unless brands take steps to exclude it is not worth it trusted sites, otherwise their ads will continue to appear on these types of sites, creating a financial incentive for them to create at scale.”
Meanwhile, Amazon’s unnamed generative AI feature, launched in September to help streamline how sellers list and describe products, began to falter in January. The user ended up entering the AI-generated text into the program, but the Amazon program failed to clear it. As a result, products ranging from lawn chairs to chests of drawers appeared in Amazon search results with names that were essentially AI error messages. “Sorry, I can’t honor this request, it violates the OpenAI usage policy,” a product name said.
The error message is an early failure and the listing has been removed. But they provide an indicator of what the online world might look like, moving away from human-generated headline bait and toward content targeted for clicks and ads but generated without human involvement or oversight.
“Internet ‘enshitification’ has become so pervasive that the term was declared the 2023 word of the year by the American Dialect Society,” Kompella noted. (The term was coined by internet activist Cory Doctorow in November 2022.)
Koppella added that this decline will intensify as AI programs learn from AI content.
google killer
The online advertising business is also facing a major shift as search engines move from a listings approach that drives traffic to websites to a snippet approach that promotes the engines’ own artificial intelligence programs. This could spark new fights.
Google, for example, started monetizing snippet search results in January by showing ads in them, but other platforms are already doing a better job with snippets. While the search giant sometimes responds with just a summary, Perplexity AI responds to each query with in-depth annotations that include links to multiple sources. Consensus does the same thing for scientific papers. The Browse for Me feature in the privacy-focused Arc Browser’s Arc Search renders a customized web page with information and relevant links while blocking ads and trackers.
“They are like personal research assistants that simplify and personalize every response,” Kompella said. “They are a game changer if people are looking for in-depth, qualitative analysis-driven responses.”
These search services will cost content creators — we can expect to see copyright and legislation-related debates over fair use of such engines soon — and it will likely cost users, Kopera said.
“The internet of the future will most likely be a walled garden space,” he added. “As automatically generated content increases, hand-crafted content will become more valuable and premium material will increasingly be available, and perhaps even visible, to paying subscribers.”
Judging by the numbers: Bigger. better one?Take a look at our online life
According to DataReportal.com, the average user spends 40% of their waking hours on the Internet. Every minute, 6.3 million searches are conducted on Google worldwide and 6,944 tips are entered into ChatGPT. Take a look at other compelling findings from the 2023 report from reference library DataReportal.com, research agency Statista and cloud solutions company Domo.
*Between 2013 and 2023, the number of global Internet users has almost doubled, and the number of user profiles has grown from 2.7 billion to 5.3 billion.
* As of January, India had 751.5 million Internet users, with an Internet penetration rate of 52.4%.
* Norway and Denmark are the most connected countries in the world, with Internet penetration rates as high as 99%. Meanwhile, millions of people remain offline in countries such as North Korea, which reportedly have the lowest internet penetration in the world at less than 10%.
* Users spend an average of 6.5 hours online every day, accounting for approximately 40% of their waking hours. DataReportal cited a 2023 survey by market research firm GlobalWebIndex as showing that seeking information, staying in touch with loved ones, and leisure/entertainment accounted for the majority of time spent online.
* The typical social media user now spends 2 hours and 23 minutes per day on these platforms, accounting for more than one-third of total online time.
* The number of Google searches is soaring around the world. There were 500,000 queries per day in 1999, 2 million queries per minute in 2013, 5.9 million queries per minute in 2022, and 6.3 million queries per minute in 2023.
*ChatGPT currently holds the record as the fastest growing consumer application in history. In January 2023, two months after launch, there were approximately 13 million unique visitors per day. That number rose to 100 million monthly active users that month. In November 2023, CEO Sam Altman announced that ChatGPT had reached 100 million weekly active users.
* Digital advertising spending has more than doubled in five years, from $331 billion in 2018 to $720 billion in 2023. Most of the spending currently is on programmatic advertising, a format that tailors ad streams to each user based on data that has been collected about their online activities, browsing history, spending habits and other such parameters.