AI is devaluing the internet, and it might be a good thing

Where have we come from in the development of the internet, and where might AI be leading us next? I don’t know, but I have some thoughts.

Early expertise created a baseline for quality

When people first began to venture onto the nascent internet, there was relatively little content. The earliest users were professionals working in scientific research, and governments. As the floodgates were opened to the masses, the tools, and platforms available to publish content still required a certain expertise. That created a limiter on who was publishing content.

The difficulty of finding content pre search engines, and really pre-Google, because the precursors were pretty awful, also meant the occasional crank spewing their manifesto onto the internet wasn’t being encountered by random people who might get drawn in.

Vox populi

Google changed this somewhat. Especially as its algorithms evolved to show you what it thinks you want to see, rather than what the most accurate and truthful information available is. Those crank turned out to be pretty engaging. Once the search engine realized that a user’s biases aligned the extreme content a site was touting, they would ram it down their throat at every opportunity because it drove engagement. If they identified user B was similar to our first user, they’d start showing them the more extreme sites too in the hope it would drive even more engagement.

Remember, search engines don’t make money by returning quality results. They make money from getting you to look at, and sometimes click on, advertisements. If displaying results full of deceptive and untruthful websites is more effectively driving your engagement with their platform, that’s what they feed you.

The advent of social media in the late ’90s, and its rise to domination in the mid 2000s filled the internet with low quality slop. Platforms that made it easy for anyone to create a website handed everyone a megaphone to scream their agenda into the void where it might find sympathetic ears. They could blast it on social media where it would get at least some eyes, any if they were lucky get shared more widely.

Unfortunately lots of people give these sites as much credence as their favourite news outlets. Some may place even more trust in these alternative sources for information.

What though will happen when the internet is so full of AI slop, with everyone, left, right, and center, producing their own endless river of propaganda?

Vox machina

While major media outlets still generally stick to the truth, spin and bias aside, many of the alt media are much looser with their facts. Even prior to AI, social media was awash with images of politicians and celebrities, quotes of things they never said superimposed over them. AI is making it trivial for anyone to create fake videos of politicians and they will become increasingly hard to detect as the AI models continue to improve.

We can legislate. We can make it illegal to produce AI videos of individuals without their consent. Though in the US this may run into 1st Amendment issues. The courts will have to thread the line between free speech and libel. Does it really matter though when we’ve already seen foreign governments operate behind the scenes campaigns to influence elections. Russian trolls don’t care if something is illegal in the USA.

What I suspect we’re going to see is an ever increasing deluge of AI generated images, videos, and text that simply becomes noise. Static on the wire. It will undermine our trust in any online sources and drive down the value of any and all content. If I can’t tell whether an image, video, news story, etc… is real and true, why bother giving it my attention in the first place? Sure there’s always room for light entertainment, but if youtube was flooded with zero effort AI generated content, would you really bother to spend time there?

Walled gardens

While an overwhelming amount of AI content will devalue content more broadly on the internet, there’s also a darker aspect to consider. AI companies are vacuuming up every piece of unique content to train their models. This changes the motivations and rewards for publishing unique content online.

Whether it’s fan fiction, artwork on sites like Deviant Art, Photos on Flickr, if you know your unique content is going to be coopted to train AI, you’re less motivated to host it publicly. I wouldn’t be surprised if we see an increase in walled gardens. Sites that allow communities to come together in spaces closed off to AI.

We already see the start of this with many subreddits banning AI generated content. I can see room for a new social network where members of your sub-group are vetted and vouched for by other people before they’re invited to join.

Legislate or validate?

Fundamentally legislation and validation rely on people to be good citizens of the internet. They’re not. Let’s say we passed a law saying news organizations cannot use AI generated images and videos. How does one prove a piece of content is AI generated? The fact gemini adds metadata enabling identification is a nothingburger. Designed to make them look good. It takes little effort for somebody to work out how to strip metadata from an image or video.

From the various plagiarism detection services being offered we have a pretty good idea AI is no better than we are at attempting to detect content generated by other AIs. There’s no particular reason to believe it would be any different for images and videos.

Using some kind of provenance system would be a nice idea. I’m sure the blockchain bros would love to try and crowbar Merkle Trees into another use case. However a blockchain doesn’t tell you whether something is AI generated or actually created by human hands. It just tells you who put it on the blockchain.

Touch grass

Ultimately I hope that this might drive all of us terminally online content consumers back into the real world. The internet is great, it’s useful for staying in touch with friends and family, booking a doctors appointment, and so many other things that AI slop simply doesn’t feature in.

Ironically I believe Facebook investing so heavily in AI might be the thing that kills their platforms by driving the value of the content into the ground. Would it be so bad if we spent more time in our real communities, and less time in online simulacrums?