YouTube has a new plan to deal with the wave of AI-generated content flooding its platform, and it affects you too. The company is now asking viewers to rate whether a video feels like it uses AI. On the surface, this sounds like a reasonable way to deal with low-quality AI content in your feed. In practice, it can cause more problems than it solves.
Humans are bad at recognizing AI-generated content, and it’s getting worse
The most fundamental problem with this approach is that humans cannot recognize AI-generated content well, and the gap between human recognition and AI capability is rapidly widening. Early AI content had obvious signs such as robotic voices, distorted hands, or unnatural-looking faces. Newer models have largely eliminated these problems.
Voices now sound natural, faces are convincing and the obvious traitors disappear. The tools have clearly evolved, but casual viewers haven’t kept up. And there is research to back this up.
A recent study on AI facial recognition found that humans performed only slightly better than chance at identifying AI-generated faces. Even more worrying, their confidence in AI face recognition was consistently higher than their actual accuracy. Research shows similar patterns elsewhere.
A study on deepfake detection found that people have difficulty detecting deepfakes but still believe they can. Research into AI-generated speech recognition suggests that AI voices are now almost indistinguishable from real ones for the average listener.
YouTube’s own track record doesn’t help this case. A Kapwing study found that around 21% of the first 500 videos recommended to a new account were classified as AI slop, while a New York Times investigation found that more than 40% of recommended short films aimed at children in a 15-minute session contained low-quality AI content.
This is content that has already passed YouTube’s automated and human review systems. When these systems allow so much AI to slip through, it seems unrealistic to expect anything better from viewers.
The rating system also opens the door to abuse
Even if viewers were reliable AI detectors, the new ratings system is vulnerable to abuse. Coordinated campaigns against creators are a well-documented problem on YouTube, with malicious actors targeting channels through mass reporting and dislike bombing. A feature that allows users to flag content as AI junk gives them a new tool to exploit. Rival channels, angry communities or organized groups could abuse it to flag videos, regardless of whether AI was actually used.
YouTube hasn’t explained how it will vet or weight these ratings, leaving plenty of room for manipulation. Creators who have spent years building their audiences may now have to contend with a new risk that has little to do with the quality of their work. If the system is widely deployed without safeguards, it could harm legitimate creators just as it targets low-quality AI content.
And what do the viewers get out of it?
Even if YouTube somehow manages to combat abuse, there is another clear problem with the system: incentives. Tagging AI content takes effort and requires a certain level of awareness about what AI tools are actually capable of. However, YouTube doesn’t offer viewers a clear advantage when it comes to detecting AI errors. The platform, on the other hand, gets a cleaner feed and a steady stream of user data without giving much in return.
There is also a legitimate concern that there is nothing stopping YouTube from using this feedback to train future AI models, potentially making AI-generated videos even more difficult to recognize. In fact, it could transform a system designed to combat AI aberrations into one that helps improve it.
YouTube’s approach misses the mark
The new rating system is another attempt by YouTube to show that it is taking the AI slop problem seriously, but the platform is still not doing enough. It does not specifically prohibit creators from publishing AI-generated content, and although disclosure is required for AI-altered or synthetic media, this rule only applies in certain cases. The monetization penalty is also limited because it relies on the same detection systems that already let too much low-quality AI content through.
YouTube has helped create the conditions for this problem by allowing and monetizing AI-generated content for years, and its efforts to curb this problem have always failed. Outsourcing cleanup to viewers, without explaining how their data will be used and without offering anything in return, treats them more like a free resource than a community. If YouTube is serious about combating AI slack, it must own the solution and not leave the task to viewers.
