AI is destroying the Internet

Jeunese Payne
2d
3 min read

The transition to prioritising AI-generated overviews over original content might feel efficient and useful, but it actually disrupts the traditional economic model of the Internet (where people search for something, decide what to click on from the results, and enter a website), which has consequences for content quality. By keeping users on the search page (instead of sending them to the websites AI claims to reference), creators and publishers are decreasingly incentivised to post new, high-quality material.

Many of the sources AI cites aren’t even real, but most users aren’t clicking through to check. Humans love a quick and easy answer, and so AI gets away with making up its content and sometimes also its sources. To provide a link that proves its own summary, AI will unashamedly do two things:

If a name or title sounds plausible in a specific context, AI will use it to create a non-existent source;
AI will force a match between a statement and a real source where a match doesn’t exist.

Additionally, because AI models are increasingly trained on existing AI outputs rather than actual human-written content, we’re at serious risk of long-term degradation of online information. You can see this visually in attempts to get AI to recreate an image exactly, with disastrous consequences. For example:

https://www.youtube.com/shorts/CTusyU_oEBE

This happens because AI doesn't 'know' facts or 'understand' what it consumes. It only predicts which words (tokens) naturally follow others, which only makes it look like it knows facts and understands content.*

The same is true for images. Just as language models reduce sentences to word tokens, an image model breaks images down into visual tokens and then outputs mathematically probable combinations of those tokens. It doesn’t recognise what’s in an image it’s given or in an image it creates.

It’s not just AI consuming AI that’s the problem.** As human writers and content creators rely more on AI to create their own content, the Internet risks becoming overrun with AI-derivative text and media, making high-quality human material even scarcer, skewed towards homogenised information that lacks depth and originality, and pushing biased content.

By converging content based on averages and patterns, AI regurgitates uninspired content. Worse, because it generates more engagement, AI recycles human prejudices repeated throughout history (racism, sexism, etc.), and prioritises emotionally charged and controversial content. This amplifies the worst of human nature, and creates dangerous feedback loops, echo chambers, and misinformation.

Further skewing the information you get, AI summaries increasingly have adverts inserted into their answers. While search results have been traditionally flagged as 'Sponsored' where appropriate, AI summaries embed paid recommendations into its answers without disclosure. Ads for specific products are disguised as neutral AI responses. Even if/when AI providers start highlighting paid ads above and below their summaries, there are no guarantees that an ad isn’t also covertly injected into AI answers themselves.

Over time, the capacity for unique, original, varied insights, and the opportunity to provide relevant context and nuance, will diminish, and with it, so will trust, creativity, and innovation.

Rant over.

Footnotes

* The bias towards thinking verbal fluency, writing ability, and reading ability equate to general intelligence existed well before LLMs seemed to mimic these skills. For example, many non-verbal and minimally verbal autists score highly on non-verbal tests of intelligence, yet the mainstream assumption is that to lack the ability to speak equates to a lack cognitive ability. As another example, where dyslexic children used to be labelled as stupid, we now recognise that dyslexia is a specific learning difference independent of general intelligence.

** No, AI didn't write that sentence. Sometimes it's just a good way to start an argument.