You may not know exactly what “slop” means when it comes to artificial intelligence, but you probably have some idea.
Slop, at least in the fast-paced world of online message boards, is becoming a broader term to describe poor or unwanted AI content on social media, in art, in books and, increasingly, in search results.
Google suggests adding a non-toxic glue to help cheese stick to pizza. That's miscellaneous. That cheap digital book that looks like what you're looking for, but isn't, is miscellaneous. That post on your Facebook feed that seems to have come out of nowhere is miscellaneous. That's miscellaneous.
The term became even more popular last month when Google incorporated the Gemini AI model into its U.S.-based search results. Rather than directing users to links, the service tries to resolve queries directly with “AI summaries,” chunks of text that appear at the top of a results page that use Gemini to guess what users are looking for.
The change was a response to Microsoft's integration of AI into Bing search results, but it quickly led to some missteps and prompted Google to say it would roll back some of its AI features until the issues were resolved.
But with major search engines prioritizing AI, it looks like for the foreseeable future we’ll likely be served up vast amounts of machine-generated, rather than primarily human-curated, information as part of everyday life on the internet.
Hence the term slop, which conjures up images of unappetizing piles of food being shoveled into animal feeders. As with this type of slop, the AI-assisted search is completed quickly, but not necessarily in a way that any critical thinker would tolerate.
Christian Hammond, director of Northwestern University's Center for Advancing Machine Intelligence Safety, pointed out a problem with the current model: the information from AI Overview is presented as a definitive answer, rather than a starting point for internet users to begin researching a particular subject.
“You search for something and it gives you what you need to think about, and it actually prompts you to think,” Hammond said. “The integration with language models is making it less of a prompt for thought and more of an prompt for acceptance, and I think that's dangerous.”
Naming a problem helps to target it, and slop is one option, but whether it will be embraced by mainstream audiences or consigned to the dustbin of slang terms like cheugy, bae and skibidi is still an open question.
Adam Alexic, a linguist and content creator who goes by the social media handle “etymologynerd,” believes slop has potential, even if it hasn't yet caught on to a wide audience, according to him.
“I think this is a great example of a word that doesn't stand out because it's a word that everyone is familiar with right now,” Aleksic said. “It's a word that just naturally applies to this situation, so it doesn't stand out as much.”
The use of “slops” to describe low-quality AI material appears to be a reaction to the release of an AI art generator in 2022. Some point to developer Simon Willison as an early adopter of the term, but Willison, who has been a driving force behind the phrase's adoption, says the word was in use long before he discovered it.
“Actually I may be quite late to the party!” he said in an email.
The term originated in 4chan, Hacker News and YouTube comment sections, where anonymous posters would sometimes use in-group language to demonstrate their mastery of complex subject matter.
“Any slang usually starts in a niche community and spreads from there,” Aleksic says. “Usually it's coolness that makes it spread, but that's not always the case. For example, a lot of words come from the programming nerd community; look at the word 'spam.' Usually words are created because there's a certain group of people who have a common interest and a common desire to invent words.”
In the short term, AI’s impact on search engines, and the internet in general, may not be as extreme as some fear.
News organizations have worried that their online audiences will decline as people become more reliant on AI-generated answers, and data from internet traffic tracking company Chartbeat shows that there was an immediate drop in referrals to their websites from Google Discover in the first few days that the AI-generated summaries were published. But that drop has since been reversed, and overall search traffic to more than 2,000 major U.S. websites actually increased in the first three weeks of the summaries' publication, Chartbeat said.
But as people become more accustomed to AI's growing role in how the internet functions, Willison, a self-described optimist about AI as long as it's used correctly, thought slop could become the go-to term for lower-quality forms of machine-generated content.
“Society needs a way to talk succinctly about modern AI, both the good and the bad,” he said. “'Ignore that email because it's spam' and 'Ignore that article because it's rubbish' are both useful lessons.”