I read a great piece recently about the prospect of AIs getting progressively less intelligent as they learn from other AIs. As AIs create text from statistical analysis of the human language, they less often output text that is less often seen in training sets. As the proportion of AI content on the internet increases, these "rarer" human responses become an even smaller portion of the AI training set, hence are output by AI even less. Studies have shown AIs learning from other AIs become increasingly biased and more simplistic. Our reliance on these increasingly less intelligent AIs in the future could work to instill yet greater biases in our own brains than we already have. Is this possibility addressed in any AI legislation ? Probably not, given there isn't enough rules governing AIs regarding the other, more catastrophic possible outcomes for humanity. Nevertheless this should be taken into consideration as legislation governing AI is drawn up. There are a lot of ways AI can help us- but better keep on top of the ways that it can harm us. Perhaps new rule: limit AI produced content in training sets?