AI Internet Litter

2023-06-17 – #Rant #AI

I write this in June 2023. ChatGPT was made public in November 2022 and ever since every other news feed is being occupied by AI news and trends and vague future predictions. It just doesn’t go away. So I will take a moment to talk about how AI already made the internet worse .

Remember how in the first wave of AI coverage, every article and YouTube video started with a generic introduction only to then disclose that the first sentences have been generated by ChatGPT 🤯. What if I were to tell you that the last paragraph has been written by ChatGPT too?

Auto-generated content has been here for a long time

The internet is already very noisy. To even be noticed, you have to value quantity over quality in some respect.

Figure 1. You surely have noticed such sites

If you search for "programmatic SEO", you will run across the early stages of generated content.

Notice how this is written by a SEO agency. They don’t think there is anything wrong with that. They will try to sell you exactly that as a service.

As an example, Tripadvisor has a page for almost everything travel related. Yelp has a page for all business searches. That’s because they use programmatic SEO to reach their target audience. If someone searches “top things to do in (city),” Tripadvisor is always in the top results.

— https://breaktheweb.agency/seo/programmatic-seo/

I notice that practice when I want to know how to make two frameworks work together. As a constructed example: Let’s say I want to know how to apply the redux pattern in the svelte framework. I might google "svelte redux" and very often one hit will be something like "reduxjs vs svelte". The comparison between reduxjs and svelte is obviously comparing apples to oranges. No real person sat down and created that site. It was auto-generated once and now clouds peoples search results for eternity.

We can even push this idea a bit more and find sites that try to do bend spacetime and compare CSS and SQL.

I went with the programmatic SEO example here to drive my point home. But even in various news sites like the german finanzen.net auto-generated articles are a standard practice. Feeding for example press releases into a simple NLP program is the easiest and cheapest way to add articles to your site. You could also feed the current stock price into a NLP program and let it comment on it every other hour.

Luckily, there is little to no AI involved in these tools. It is blatantly auto-generated. You maybe chuckle a bit when you stumble across one of these and go on with your life. Maybe some news articles end up sounding very similar to their competition. Only minor inconveniences.

Spicing things up with AI

One YouTuber I enjoy watching is "Geschichtsfenster". He is a german YouTuber who got famous by reacting to TV documentaries on the middle ages and explaining how they are often really bad and factually wrong. Whenever these documentaries depict shit-flooden streets, dirty and unfriendly everyday people, the subtle gray-brown "middle age" video filter or even just clothing that is inappropriate for the time, he is the first one to it point out. The reputation of the middle ages has been killed by documentaries like that and he is fighting the uphill battle of straightening that out a bit.

One merch article of his is the "allmächtige Kirsche" which translates into the "allmighty cherry" (depicting just that) but in german that can be pronounced as the "allmighty church" too. This makes fun of every other documentary that will state that the church was having their say in everything all the time.

Now he got confronted with an AI-generated video on the great famine of 1315. Of course we can’t know for sure whether the audio and the text are AI-generated too but the image slides surely are. And when the visuals of the video are AI-generated what reason is there to believe that the script isn’t copy-pasted from ChatGPT as well? The voice was well made but it might also just be text-to-speech. Maybe the rest of the video is real and human-made but solely the idea that it might not be is enough. We are at a point where not only text articles, but whole videos can be AI-generated.

Humans in old clothing bow over a skeleton on the floor

Figure 2. AI-generated image on the famine of 1315, which might just as well be a primary source from the middle ages

What we are now seeing is the rise of such AI-generated content. The difference to the already prevalent auto-generated content is that you don’t know it’s computer-made anymore.

Clues on AI-generated content (the paranoid bit)

Now, I have become a little paranoid about this. I now often become suspicious of content I consume if there is a chance it is AI-generated. Then there is a sense of relieve when I stumble across:

minor spelling or grammar mistekes
directly relevant diagrams and images
very contemporary pop-cultural reference
direct and verifiable quotes or links

All of which reduce the chance that it is AI-generated but they are not a guarantee. Images are generally a good sign that content is not AI-generated except if the images are only very vaguely associated with the topic. If the topic of the article is hiking in the alps and the image looks like a stock photo, it might be AI-generated. If the image depicts a specific route through the alps or a height chart, it is most likely not auto-generated.

And maybe the pop-cultural references are one solution. We just all need to get into the habit of referencing short-lived internet phenomenons to attest our authenticity. And what better way to show that you are human than to reflect some culture.

Figure 3. Holding up a newspaper to validate your humanity in the face of auto-generated content

But of course all these clues will eventually fade away as AI gets more advanced.

Conclusion

The internet is full of garbage (👋) and it gets worse every day. This is not surprising, it is just that AI will get us there faster. All clues to distinguish AI-generated content from human-made content will slowly fade away, just as they did for auto-generated content. If you don’t want to end up in a situation in which half the content you consume on the internet is AI-generated, then we need to come up with a system to fight against this.

One way to do that would be to subscribe to my RSS feed.