AI slop has broken the internet

"Slop" was Mirriam-Webster's Word of the Year last year. "Slop" or "AI slop" is defined as "digital content of low quality that is produced, usually in quantity, by means of artificial intelligence". You know the kind of thing: pictures of Shrimp Jesus, glue-topped pizza recipes, AI-generated songs, sham biographies, deep fake videos, talking cats, genuinely fake news, advertising that looks "not quite right". AI slop is everywhere, and it now, after just a few short years, makes up a good proportion of the entire Internet. And for this we are investing billions of dollars, and straining our electricity systems to power endless miles of data.centres.

I knew it was a major problem, but I hadn't appreciated just how much of a problem until I read an article in the Globe and Mail (originally from the New York Times) today. It seems that tech giants like Google, Meta and X are having to spend significant resources weeding out AI slop that was largely produced by their own platforms.

Spotify removed 75 million AI fake songs and duplicate uploads last year. Google wiped 130,000 YouTube channels of low quality content over the last six months. TikTok removed 377,000 videos in the first three months of the year for violating its AI policies. (TikTok USA is developing a feature that let's people dial up or down the amount of visible AI content.) 

But still the problem lingers and grows. Apple Music estimates that a third of what is uploaded to its site is now AI-generated. French streamer Deezer says it may be as much as half in its case. More than 20% of short YouTube videos are AI-generated slop according to one (AI-powered) online video editor. Nearly half of the posts over 50 words on X are create by AI.

But the same tech giants that are trying to manage AI slop are continuing to make things worse. At the same tume as Meta showcased a detection, labelling and tagging system designed to make AI content more identifiable, it also rolled out an all-AI video stream widely considered to be the apotheosis of AI slop, and released an AI image-generator using its own public Instagram accounts as its references.

Part of the problem is that the less-discerning elements of the internet's clientele laps this stuff up, and AI content, however poor, can substantially inflate the viewership of many platforms. It can make big money for the creators and re-posters too. Unfortunately, it also makes it much harder for other users to find genuine, truthful and high quality search results.

As Substack's CEO put it, "it's getting harder to tell what's real on the internet", and "platforms that reward fakeness will create a race to the bottom". Quite. Substack recently announced its own AI detection tool for users, but the tool mistakenly identified some real, human posts as AI-generated.

This stuff is not easy. The point is, though, the onus should not be on users to figure out what's real and what's not. It seems trite to whine that "the internet is broken". But, really, the internet is broken.