Just Like Humans... Artificial Intelligence May Suffer from 'Brain Rot'

SadaNews - Over the last few years, the term 'brain rot' has spread among scientists and health experts, referring to a condition that affects humans who experience short attention spans, distorted memories, and increased mental health issues such as stress and depression due to overconsumption of content considered trivial or of low quality, especially on the internet.

Researchers now say that the same phenomenon can also affect artificial intelligence models, according to a report by 'Fortune'.

In AI models, continuous exposure to short and widely disseminated social media posts leads to 'permanent cognitive degradation in large language models', according to a new study that has yet to be published.

To test their hypothesis, researchers continuously fed large language models with short and widely disseminated posts on the platform 'X'.

They found that 'this toxic training causes a noticeable decline in the ability to think and understand long contexts, partly due to an increase in (idea skipping), meaning that AI models increasingly fail to formulate a plan to answer questions, overlook parts of their logical analysis process, or completely bypass and ignore the process altogether'.

When researchers attempted to 'treat' large language models using high-quality human-written data, they found that the models still suffered from residual effects of cognitive degradation, showing a significant gap in the quality of their thinking before and after training on social media posts.

The researchers stated that this gap indicates that the effects of brain rot have been deeply ingrained, and the degradation in their performance may be long-term.

As artificial intelligence models are trained on trillions of data points from across the internet, researchers warned that they are 'inevitably and continuously' exposed to this low-quality content just like humans, which could pose risks to the technology as a whole.