Will AI models collapse when the majority of web content becomes AI-generated?
AI models are getting better at creating content and humans are getting better at asking them to create more content and in some cases, agents are creating content and publishing without human edits or inputs, autonomously.
To those these agents are serving, this increased capability of creating good content is progress, but when enough people get into the game of relying on AI for this, what happens?
Well first of all, the Internet becomes heavily dominated by AI-generated content.
A lot of people out there refer to AI content as slop, which generally means that they are not valuable and effectively trash.
So if this definition holds weight, or measures right, to even the slightest considerable percent, then we'd be left with a situation where the Internet becomes a trash pile.
Fun fact: the Internet is a trash pile
This is a reality most people miss and it's the reason I'm willing to discuss this topic after seeing it around.
The Internet was always a trash pile, only with some pretty huge volume of valuable things.
But the reason you never actually had to deal with the trash, which was the majority, was because of algorithms that ensure you only saw what would keep you coming back.
You don't see much of the spam or trash content that actually gets put out because, well, the algorithm has learned that it's cat videos that bring you back online after 2 mins of logging out.
Of course, AI amplifiers the rate at which trash can actually be put out.
Far more content can be published today by just anyone even without complete automation of the process, so imagine the pace and volume when said automation is in the picture.
Is a model collapse inevitable?
One thing people seem to always forget is that the intelligence of machines isn't measured against a tree’s branch in the forest, but humans.
What this should tell anyone is that a model's capabilities will generally always be about impressing the humans.
This means that before a model collapse would occur in the scale often thought of when this topic is being discussed, AI would have made necessary adjustments or improvements to meet human needs where they are.
If people don't hate the majority AI content on the Internet for some reason, most people won't recognize AI models to have collapsed in capabilities.
Besides, the Internet will not be the only source of data. Video feeds, audio, all of these will come to play at a much bigger scale than today.
There's the obvious privacy concerns here, but that's not the focus of this post.
AI is more likely to run into improvement issues related to costs of building the needed systems to help it improve than it is to run into data problems.
Posted Using INLEO