If you've spent much time on TikTok recently, you may have noticed a strange new type of AI 'brain rot' taking over: fruit dramas. These AI-generated short dramas feature odd-looking anthropomorphic fruit characters engaging in ethically problematic behaviours, including affairs, racist attitudes, and sexual assault of women characters. One account, ai.cinema021, which has launched a parody series called Fruit Love Island, has amassed more than 3 million followers.
At face value, the videos come across as bizarre and grotesque, but they are amassing hundreds of millions of views. This content is, at best, a water-guzzling affront to animation and, at worst, actively normalising racism and misogyny. So why does it have so many fans? These videos exploit core features of human psychology, combined with addictive platform features such as infinite scroll, resulting in an endless stream of content that keeps us engaged.
Short-form video feeds operate on principles similar to gambling systems. The human brain is highly sensitive to novelty and unpredictability, both linked to dopamine signalling in reward learning. When rewards are delivered unpredictably, behaviour becomes more persistent—a pattern known as 'variable reinforcement'. AI slop videos offer rapid visual novelty and unexpected emotional turns, compressing big emotional experiences like betrayal, sadness, revenge, and humour into seconds.
Many viewers describe a sense that these videos feel 'off', relating to the uncanny valley where near-human representations produce discomfort. However, they rarely become disturbing enough to trigger avoidance, instead sitting in a middle zone that provokes curiosity. This creates cognitive tension, and according to cognitive dissonance theory, people are motivated to resolve such inconsistencies by continuing to watch in search of closure.
We are also more likely to ignore unethical messaging because the highly synthetic characters make scenarios feel fictional, even when they reflect real social behaviours. Research on moral disengagement shows people relax ethical judgment when harm appears abstract or indirect. Social media algorithms do not prioritise meaning or quality; they prioritise content that captures attention, driven by metrics like watch time and completion rate, creating a feedback loop.
From an AI governance perspective, these videos highlight an overlooked risk: generative systems can gradually shape our behaviours without us realising. Avoiding social media entirely is not realistic for many, but small changes can reduce the pull of AI-generated brain rot. Introducing a pause before scrolling to the next video can weaken the reward loop, making it easier to put the phone down.
