Why internet searches are serving up 'AI slop'

These days it feels like hardly a news bulletin goes by without some sort of mention of AI technology and its power to do things in seconds that might have taken us hours without it.

But our understanding of how it does those things still is not great, by and large, and there are still the things that it is doing which we simply are not aware of.

Meanwhile, so-called "AI slop" is coursing through the internet all over the place - unwanted, unloved and perhaps even unstoppable.

Danyl McLauchlan, author and IT systems consultant at Victoria University of Wellington's School of Biological Sciences, said science fiction has foretold what happens when such content is left to spread unchecked.

"Neal Stephenson… describes nations in the future as having been 'Facebooked', which is… that kind of phenomenon of being completely swamped by meaningless gibberish," he told RNZ's Saturday Morning.

Stephenson coined the phrase 'metaverse' in 1992's Snow Crash, which directly inspired the name of Facebook's parent company, Meta - which has poured billions of dollars into creating the kind of virtual reality Stephenson wrote about in his dystopic novel. Facebook has in the past been accused of creating algorithmic 'echo chambers', in which people are served up engaging content which might differ completely from what others are seeing.

"There is no coherent media or what he calls 'consensual reality'," McLauchlan said. "No one agrees on what basically is happening in the country because everyone is getting different algorithmically sort of news feeds.

"And so the nation collapses, because you do need that kind of consensual reality to even argue meaningfully about what's going on and what should happen."

Another example he cited was Jorge Luis Borges' The Library of Babel.

"It's my favourite short story, so I'm really happy that people are now being terrified by it. But it describes a universe that is just one vast library and every book that can possibly be written is in this library.

"And so in theory the answers to all of the meaningful questions in life are hidden in the library somewhere. But because the vast majority of books that could possibly be written just contain strings of, you know, of alphanumeric characters that have no relation to each other, they're just filled with gibberish.

"You would need to sort of like you know, cross trillions of human lifetimes just reading through the books over and over again to find a single coherent sentence, let alone anything meaningful and true. And so that is the comparison that's often made to the torrents of AI slop that are taking over the internet."

pic.twitter.com/H0ExcNXBdl

— Donald J. Trump (@realDonaldTrump) August 18, 2024

Rise of the machines

The pace of development in generative AI has skyrocketed in the past few years. It began in earnest in the late 2010s following the publication of a paper called 'Attention Is All You Need' by eight Google scientists, which introduced a new machine learning technology known as a 'transformer'.

Generative AI broke into the wider public consciousness with the release of Open AI's ChatGPT in late 2022 ('GPT' stands for 'generative pre-trained transformer').

Whereas before generative AI was a curiosity and only accessible to those with significant computer knowhow, now it could be used by anyone with an internet connection - as easy as making a Google search.

Since then, generative AI has only gotten better - and is now able to make images, music and video too.

Perhaps inevitably, some found a way to make a quick buck out of it.

"If you go to Amazon and search for a reasonably popular book that you've read recently and I read - well, I'm still reading Rory Stewart's book Politics On The Edge, which is a sort of political best-seller in the UK - you search for that and you get Rory Stewart's books, and he's written a number," McLauchlan said.

"But you also get these quite strange results which have titles like Rory Stewart Book and Rory Stewart Political Career and they're presented as books… Rory Stewart Book you can purchase online for $7.89.

"And if you look at the reviews, you'll see that there are only one or two and they're completely outraged because they have [received] a book that is not written by an ex-politician or a human. They're written by a generative AI, and they're basically just gibberish.

"They've taken the text of one of Stewart's books and rewritten it so that he can't sue for plagiarism and then sold it as an online text. And so it's presenting itself as something meaningful, but it's actually just generated algorithmically and it's completely meaningless - and that's the case for just about anything that's reasonably popular."

Even kids books are being plagiarised via computer, he said.

"You can do this in a lot of different domains. You can do it with recipes, you can do it with celebrities. Google is starting to crack down, so you do need to go a couple of search pages deep. TikTok and Instagram are not cracking down, and so you'll be very quickly presented with just the most inane meaningless gibberish you can imagine.

"I was looking at celebrities - Donald Trump, Joe Rogan - and on TikTok and Instagram, you'll find these so-called 'stories' and they have they have deepfake voices of these people saying things. And you know, we've all been primed to think, 'Oh, the Russians are creating this, you know, deep fake technology to destroy our democracy.' But actually, it's not political content, it's not even content. It's just random gibberish in the voices of these celebrities.

"And so what they're doing is they're creating vast volumes of this content as a form of search engine optimisation. So, whenever anything searches anything, you'll increasingly get AI slop… you can use these open-source models and just spin up a instance on a cloud computer and generate gigantic volumes of garbage."

Grok AI is allowing users to make images with public figures, copyrighted characters.

Here's what I generated in few seconds:

1/ Donald Trump and Kamala Harris pic.twitter.com/urf3JU1Nbo

— Nitish Gujjar (@Nitish_Gujjar_) August 19, 2024

Attack of the clones

And that could have dire consequences for real creators, including the news media - even outlets who avoid using AI. Google is now offering AI-generated search results, so the user does not have to click through to an article to read it - instead getting an AI-generated summary.

"That just sent a shock of fear through media companies around the world because so much of their online revenue is ad revenue that's generated from Google clickthroughs," McLauchlan said.

"So all of that money will disappear from an industry that is already going bankrupt quite quickly. Google made about half a trillion dollars last year, in New Zealand dollars, and the New York Times made about 1 percent of that - and that's the most profitable news media company, at least in print media. So that sort of gives you an idea of the scale of the two competing organisations."

He was not sure what the solution might be, but had an idea where to start looking - the sci-fi which inspired the technology in the first place.

"It's the meaningful literary form of our time."