There's a post going around that is claiming that AI has overtaken human writing on the internet. Its central claim is that this is because ChatGPT and a phenomenon that started in mid-2022. Less commonly reported is that the post is PR by a company that wants to use the image of inevitability to sell you AI.
I'm something of a the-internet-is-fake expert myself so I can pretty confidently tell you that this article (PDF copy here) is bullshit in a number of different ways.
Look, if you're not an expert you wouldn't know, but this chart is clearly incorrect. Here's how I can tell: they could not possibly measure this at all right now. Why?
- You cannot tell for sure that content is AI generated. There's no reliable automated way to detect AI generated text, as has been noted by multiple publications.
- You're not counting the many sites that are blocking crawlers now.
But there's an even more important reason...
Automated text generation has existed for decades. "Dead internet" is a conspiracy that got popular back in 2016. Markov chain text generation dates to at least 1984. The incentives to algorithmically generate web text have been around since before 2010. Huge swaths of the internet have been made-for-seo/advertising slop for decades; either using mediocre text generation models or just low paid humans. NYMag in 2018: "Studies generally suggest that, year after year, less than 60 percent of web traffic is human."
This is inevitability propaganda, by an AI company named "Graphite", made to make you feel like you must use AI. It's bullshit! This methodology is bullshit, for many reasons, not least of which is: it is reliant on Common Crawl, now blocked by many of the biggest content producers online
If this was a real study it would be published in a real journal not a fking press release on the AI company's site. They don't even have the confidence in their theory to do what a lot of other AI companies do and commission real university scientists to run it. Or put it in a preprint!
Even the supplied dataset is BS and they simply do not provide much of the information that would be needed for reproducing and verifying their work.
The people who wrote this study are, at best, fully incompetent! "Since the large-scale adoption of AI tools began with ChatGPT, we argue that, with high probability, articles published before the release of ChatGPT were written by humans"
No algorithmically generated content existed on the internet before ChatGPT? Lol
'Almost 100% of the web is published by humans in 2020' is a statement that is so incredibly wrong that it is entirely invalidating of everything else this article could say just by itself!
Do not write about or share this study! Don't give these AI bullshit artists what they want.
Want reliable sources about AI? I've collected them in a context page.