Researchers: Artificial intelligence systems may be on the verge of collapse

Researchers recently warned that artificial intelligence systems may collapse into "nonsense," as more websites are filled with content created by artificial intelligence itself.

Recent years have seen increased excitement around text generation systems such as ChatGPT from OpenAI. This excitement has prompted many to post on blogs and platforms created by these systems, as more online content is produced by artificial intelligence.

However, many companies producing these systems use text taken from the Internet to train them. This could lead to a loop in which the AI systems used to produce these texts are trained on the same content.

This could quickly lead to these AI tools falling into "nonsense," the researchers warn in a new paper. Their warnings come amid more general concern about the "dead Internet theory," which suggests that more of the Internet is becoming automated in what could be a vicious cycle.

It only takes a few cycles of generating and then training this content for these systems to be able to produce “nonsense,” according to the research.

The scientists found that one system they tested using text about medieval architecture needed only 9 generations before the output was just a repetitive list of “nonsense,” for example.

The concept of AI being trained on data sets that were also generated by AI is referred to as “model collapse.” Researchers warn that this may become increasingly common as AI systems are used more online.

Source: AA